Browsing by Author "Thambawita, D.R.V.L.B."
Now showing 1 - 14 of 14
Results Per Page
Sort Options
Item Android mobile driving assistant for highway drivers(Uva Wellassa University of Sri Lanka, 2015) Iloshini, P.A.A.; Thambawita, D.R.V.L.B.Mobile based driving assistant that always communicate with the driver in an interactive way, has not introduced yet. It assists the driver when he is driving on highways and driver can control the assistant through his voice command. Mobile Driving Assistant helps driver to clarify the way that he drives. In addition, the driving assistant carefully examines the facial expressions of the driver and if he is in a drowsy condition, assistant suggests alternatives to overcome from those situations. This is a friendly interactive guide for the highway drivers. Recalling the history, in 2002 Ji and Yang (2002) has presented a detection drowsiness system based on infrared light illumination and stereo vision. This system localizes the eye position using image differences based on the bright pupil effect. Brandt et al. (2004) has shown a system that monitors the driver fatigue and inattention. For this task, he has used VJ method to detect the driver’s face. Using the optical flow algorithm over eyes and head this system is able to compute the driver state. Tian and Qin (2005) have built a system for verifying the driver’s eye state. Their system uses Cb and Cr components of the YcbCr color space; with vertical projection function this system localizes the face region and with horizontal projection function it localizes the eye region. Once the eyes are localized the system computes eye state using a complexity function. Pallavi M, S. Gawali in 2012 their research paper demonstrated the new non-intrusive approach for monitoring driver drowsiness depending on the driver and driving data fusion. They use percentage of eye closure (PERCLOS) model for estimating driver status. The driving information such as lateral position and steering wheel angle also use for drowsiness detection. Multilayer perceptron neural network has been trained for optimal performance score in this research paper. Yong Du, Peijun Ma in 2008 published a research paper on effective vision based driver fatigue detection method. In this at primary stage, the inter-frame difference approach binding color information is used to detect face. Marco Javier Flores and Jose Maria Armingol in 2008 presented the basic model for drowsiness detection. For this they used Viola & Jones (VJ) method to detect the driver’s face. Once face is detected SVM is used to detect eye status from trained data. Methodology For implementation of the Mobile Driving Assistant application, Samsung Galaxy Core was used as the mobile device and the android version 4.1(API level 16) was used as the development environment. The internet and GPS services need to be activated in Mobile Device. The Mobile Driving Assistant is based on android platform supported mobile phones only. Java was used as programming language and common programming language to develop android applications. ADT bundle was handled as IDE for the implementation. Android voice recognition and android Text-To-Speech facilities were focused in order to maintain the voice discussion between the driver and the mobile driving assistant. Applications that available in Android platform can potentially make use of any speech recognition service on the device that's registered to receive a Recognizer Intent. Google's Voice Search application, which is pre-installed on many Android devices, responds to a Recognizer Intent by displaying the "Speak now" dialog and streaming audio to Google's servers. The Android platform includes a Text-to-Speech (TTS) capability. Also known as "speech synthesis", TTS enables an Android device to "speak" text in various languages. Face and eye blinking detection is the most important module of the mobile driving assistant. Haarcascade_lefteye_2splits.xml files, distributed with OpenCV package were used to detect eyes when eyes are opened. OpenCV 2.4.9.0 was used for the image processing purposes.Item Automated Dispensary Mechanism for Government Hospitals(Uva Wellassa University of Sri Lanka, 2016) Rupasinghe, C. J.; Thambawita, D.R.V.L.B.Dispensing medications in a hospital is a time consuming operation. In usual practice, the pharmacist need to dispense tablets and capsules with a simple tray and spatula. A typical hospital dispensary has only a few number of pharmacists causing long patient queues, the situation can be commonly experienced in most of the government hospitals in Sri Lanka. This inspired to introduce the concept of automated medication dispensing for Sri Lanka. This research introduces an automated solution for medication dispensing in outpatient dispensaries. The key objective of the study was to reduce the time consumption of drug dispensing. A desktop application was developed to generate prescriptions and an automated drug dispensing has been implemented. The solution comprised of a software artifact and an electronic hardware unit. The software tool is connected with a fingerprint reader which is used for patient registration. Patients can register with the system when he or she comes to the outpatient dispensary at first time. The pill container has three main parts: initially the upper part of the pill container is fed with tablets to be counted, and the middle part for hold some amount of pills to fetch the rotating pill catcher. Then the lower part called rotating pill catcher for catch pills one by one and carry them to the exit gate of the pill container. This pill catcher rotate with the help of a stepper motor. An infra-red sensor keeps counting the number of pills which are being released through the exit gate. If the count is equal to number ofpills in the prescription, the motored mechanism will stop. Furthermore, two mechanisms were adopted for the tablets counting process. The first mechanism is based on a vertically rotating wheel which have several openings to catch one pill at one time. The second mechanism is based on a horizontally rotating round plate which have openings to catch one pill at one time. The process which is based on the horizontally rotation plate has achieved a considerably accurate pill count by compared to the vertically rotating wheel version. It was noticed that the vertically rotating wheel model was blocked most of the times because it cannot hold exact one pill at a time. Therefore horizontally rotation plate can be introduced for as an effective pill counting mechanism. Keywords: Pill catcher, Pill fetcher, Infra-red sensorItem A Conceptual Model for Internet Based Carpooling Systems via a Visibility Assessment to Match the Requirements of Sri Lankan Community(Uva Wellassa University of Sri Lanka, 2016) Arnolda, H.N.; Thambawita, D.R.V.L.B.Increasing traveling needs of the country has created a gigantic increase in a number of vehicles used. This has created negative effects on many aspects such as trade deficit increase, the increase in oil consumption, environmental pollution and heavy traffic congestion. Even though carpooling is considered as a well-known alternative to reduce congestion and pollution on roads its popularity is comparatively low. This research focusses on identifying a communication model that allow carpoolers to interact resolving failures in current systems with a proper incentive distribution. A questionnaire was conducted to clarify the social concerns involved in carpooling. The results were used to analyze the factors that limit users from using carpooling and to specifically identify the features that need to be inserted to the system. An integration of site's review system and peer-to-peer marketplaces which are widely used in sharing economies was used as the new model for internet based carpooling systems. Each signed up user has their own profile and the system provides many options to edit and maintain the profile these include basic information, photos, preferences, interests etc. All the users can enter vehicles to the system and it allow users to maintain vehicle profiles by inserting photos descriptions etc. Passengers looking for rides search for tours scheduled and vehicles looking for passengers can add the tour to the system. Each request passes through four stages. They are requesting, replying and pre-approving or pre-approving and confirming. After each agreed transaction, all parties can review and rate each other. This methodology for carpooling systems is more sophisticated and realistic than earlier research on carpooling modeling. It introduces internet based carpooling and interacting platform to eliminate identified limitations of current systems such as social concerns and strict schedules by inheriting methodologies used in successful sharing economy companies such as "Airbnb" and site review systems to provide better understanding between interacting parties. Keywords: Sharing economy, Site's reviewing system, Carpooling, Peer-to-peer marketplacesItem Consumer mobile community(Uva Wellassa University of Sri Lanka, 2015) Kumara, I.H.M.S.; Thambawita, D.R.V.L.B.Whole population in the world can be identified as a group of consumers. The consumer protection is one of the important issues in this global village. Most countries use different methods and concepts to protect the consumer. Empowering the consumer with knowledge is one of the effective methods to protect consumer from market frauds. The consumer community, society, and consumer education programs are used to improve consumer knowledge in worldwide. The Consumer Affairs Authority Act, No 09 of 2003 provides provision for consumer protection in Sri Lanka (Consumer Affairs Authority-2014). The consumer watch android mobile application and the Consumer Affairs Authority official web system have been published by the Consumer Affairs Authority in order to improve the knowledge of Sri Lankan consumer. The consumer watch android application provides facilities to search market prices, search consumer prices, search wholesale prices and search shopping lists in Sri Lanka. Using existing android application, consumer can only update the knowledge regarding the price list (Zmessenger - The Consumer Watch). Though these improvements are introduced and established, still many market frauds and violence are reported in Sri Lanka. Therefore this project was carried out by the consumer mobile community android application to protect the Sri Lankan consumer from specific market violence, to empower customer with knowledge, make consumer mobile community application and to improve consumer’s social awareness. Methodology The project was carried out at Uva Wellassa University using its computer facilities. Relevant consumer protection process, regulation and the technologies were collected. Android version 4.0.4, PHP, HTML, Java script, Json, SQLITE and XAMPP Control Panel 3.1.0 were used as development tools (Android Training API Guides; JSON Tutorial & AJAX Tutorial; Stack Overflow). The Use case diagram, Data Flow Diagram and the Entity Relationship Diagram were done under the logical design of the system. Android and the web development interface and the databases were developed under the physical design. Supportive admin panel web system was developed in the first stage of the project. The web system front end was developed using HTML and CSS. Ajax and the JavaScript function were used for the validation and PHP function call. Android web services and the web back end function were developed using PHP language. Android mobile application was developed in the second stage of the project. APP interfaces were designed based on responsive Xml design. Therefore the mobile application is compatible with the different screen sizes of the mobile devices. Mobile application used HTTP request and response to communicates with the web system. Json array Format was used to send and received the information between mobile application and the web site. All back end functions in the website were developed using the PHP scripts. Both PHP and the android language were provided by Json encode and the decode function. Therefore it was great technical advantage to share large amount of information between server and the mobile application. The main functions flow of the system is explained by Figure: 01 Data Flow Diagram .The consumer and the web admin are the main two users who interacted with the system. A consumer’s mobile phone is registered automatically when the application icon is clicked, after which the consumer’s profile information will be updated with the system. The login could be done after the profile is updated. After login process, consumer is able to perform functions such as make a call, search a profile and price list, share or read market frauds, make comments on share market frauds, browse web, and search GPS location etc. The web administrator can directly login to the system via web admin panel. He has permission to manage the consumer registering profile and the consumer shared market frauds. Also he can enter, update and delete data used by the consumer via mobile application .Also market fraud’s analytical reports can also be generated via web system.Item Expert food analysis system(Uva Wellassa University of Sri Lanka, 2015) Seneviratne, S.M.U.C.; Thambawita, D.R.V.L.B.Diet is a main factor which directly affects the health condition of a living being. Most of the times, unsuitable diet practices are the main reasons for long term diseases in the human body. For examples, unsuitable diet practice is the reason for occurring and increasing the diseases like diabetes, high blood pressure, heart diseases, cancers, allergies, etc. By choosing the correct diet can not only save the body from numerous diseases but also cure existing diseases. Suitability or unsuitability of a diet depend on numerous factors like existing diseases of the body, age, gender, special body events like pregnancy, season of the year etc. Apart from the human health it may be required to select a suitable diet based on the religion and cultural facts of each person. To identify a suitable diet (food products), it is required to analyze each and every food component which is used to produce a complete food product. But, due to the complexity of the formation of the food products in the current market and the less knowledge in the food analyzing domain makes it difficult to select the suitable food product base on the above mentioned facts. With the help of Information Technology and knowledge of Ayurvedic Medicine, it is possible to identify a path to reduce the difficulty in analyzing food products (Alwis, R. 2012; Hemachandra, D. and Karunaratne, H. 2007). Ayurvedic medicine is rich in knowledge to analyse a food product but it is difficult to use that knowledge by the general public due to less understanding of the Ayurvedic food analyzing knowledge and less availability of the information from the food producers regarding the food product (Nagodawithana, P. 2007). With the help of the Information Technology, it is possible to identify a pathway to provide a solution for above mentioned problems by creating an efficient and simple way for the proper flow of information among human experts in food analyzing domain, food producers and customers. By using this solution, it is possible to keep the people away from the diseases which could be occurred due to unsuitable diet selections. It will improve the life quality of the human society. Additionally it will provide a high value to the economy. “Expert Food Analysis System” will help people by minimizing the complexity when they are going to select a suitable diet based on their physical body and social status. Selection of a suitable diet will result a healthier society which is free from most of highly spread diseases like diabetes, high blood pressure, heart diseases etc. “Expert Food Analysis System” creates new business opportunities in the market. It will create a competition among food products in terms of their healthiness. That competition will keep the healthy food products in the market and will create new opportunities for healthy new food products in the food market. And also will remove the food products which are obvious reasons for diseases. The final outcome of this solution or the final goal of this product is improving the quality of life in the society by combining medicine and business with information technology. Methodology The whole project consists of three main parts as Expert System, Web Application and Mobile application. Expert System is implemented with CLIPS (C Language Integrated Production System) and is integrated into Web application using Mommosoft library (CLIPSNet; CLIPS). Web application is implemented with HTML, Javascript, CSS and C# in the Visual Studio 2010 environment. Databases were created with Ms SQL (MSDN Library; Newest Questions). Considering the whole system, Expert System can be considered as the back-end. Mobile application is the front-end and web application acts as the intermediate component between mobile and expert applications. Web application is hosted in an IIS testing server. Mobile application is developed with Java on Android platform in the Eclipse environment. Mobile application and Web application is linked together with the internet. Human food analysis experts should store their knowledge related to food analysis domain as raw data in the knowledge base of the Expert System. Raw data about various food products including ingredients should be stored in a SQL database. Web interfaces have been provided for ease of data insertion. System generates a QR code which is included an identification number which can uniquely identifies each registered food product. Generated QR code should be attached with the package of the food product or racks etc. Food consumers have to maintain a personal profile including the information related to personal health and culture. Food consumers can scan the QR code with the mobile application and validate the food product by comparing the information already printed on the package of the food product and received information by the mobile application. When a QR code is scanned, mobile application send an identification number to the web server and refer the SQL database for food ingredients and personal profile. Then those data will be analysed and send the report to the consumer which indicates how the particular food product personally will be effected. SQL database also stores the consumer feedbacks and raw data which are statistically valuable. Data transferring between the mobile application and remote server will be handled by http requests and responses.Item Low Cost, User-Friendly, Integrated Shopping Cart System to Motivate Usage of Smart Shopping Carts in Retail Industries(Uva Wellassa University of Sri Lanka, 2019-02) Pathirana, W.P.I.M.; Senarath, S.P.J.H.; Siriwardane, L.A.I.U.; Thambawita, D.R.V.L.B.The main goal of retail industries today is to provide a comfortable shopping experience to their customers overcoming the traditional ways of doing shopping. This paper addresses an approach to this fact by implementing a low cost, user-friendly and integrated shopping cart which can be used in the retail industry to provide a better shopping experience for the customer. This system comprises an automated shopping cart, the smartphone which is used as the display device and a website. The system is best suitable for supermarkets where instead of customers having to wait in long queues to check out items, they can pay for the items at the cart using a debit or credit card. The website enables the customer to create shopping lists wherever they are and use it while shopping. The web interface attached to the cart shows the customer’s current position, information of the product and the shortest path to the products and it was achieved by using the Dijkstra algorithm which was implemented using the Hipster Java Library. A model layout, similar to the shopping mall, was considered and measurements were taken of that. These measurements were broke down to similar points proportionate to the actual location points and by using the Dijkstra algorithm, the actual shopping cart locations were identified. The website is used by two main users, the admin and the customer. The admin can change the data on the website and data related to the android application. The customers can retrieve the information of the shopping mall. Indoor navigation of the cart was implemented using Bluetooth beacons. Coverage of three Bluetooth beacons was used to calculate the location. Through this research, low-cost methods to develop a smart shopping cart was identified and it was made user-friendly by implementing an easy to use website. This research can be helpful in identifying inexpensive and handy technologies to improve the usage of smart shopping carts in the retail industries.Item Optimization of Rabin Karp Pattern Matching Algorithm Based on Parallel Computing Techniques for DNA Sequence Analysis(Uva Wellassa University of Sri Lanka, 2019-02) Anjalee, M.G.M.; Fernando, W.P.U.; Thambawita, D.R.V.L.B.String matching algorithms are used to discover the occurrences of a defined pattern in a given text or a pool of strings which is widely used in detecting plagiarism, spam filtering and most importantly in computational biology including DNA sequencing. The existence and the intensity of a muted sequence in DNA caused for various diseases can be identified using Rabin Karp string matching algorithm. The main contribution of the study is to bring an efficient version of Rabin Karp algorithm by minimizing the spurious hits while using both Central Processing Unit (CPU) parallel techniques and General Purpose Graphics Processing Unit (GPGPU) parallel techniques specifically for DNA sequence analysis. The improved Rabin Karp is implemented using C language with POSIX Threads library, OpenMP and MPI and using Compute Unified Device Architecture (CUDA). When accelerating computations based on GPU, a special consideration has given to global memory, shared memory and texture memory, the types of memories with particular importance offered in CUDA architecture. By experimental studies, we investigated a new method to eliminate brute force matching and the GPU optimization is presented with stencil method ensuring efficiency in terms of memory overhead due to redundant data access in the serial CPU implementation. We have compared these parallel implementations for evaluating the effect of varying number of threads per block as well as varying DNA file sizes. The results obtained in this study present that the proposed implementation provides acceleration surpassing 36x speedup for string size 220 characters compared to a sequential (CPU) implementation. Eventually, using the empirical results, we could conclude that the improved CUDA C implementation of shared memory version can achieve 35 times of performance than serial implementation for a large pool of DNA data in string matching.Item Person Re-Identification and Tracking for Surveillance Camera Systems(Uva Wellassa University of Sri Lanka, 2019-02) Samarathunga, S.H.J.N.; Wijesinghe, M.U.M.G.C.P.K.; Kumara, R.P.S.R.; Thambawita, D.R.V.L.B.Person re-identification is a contemporary trend of technology, which is used to identify identities of people over a network of the camera system. Most person re-identification systems are working based on the colour histogram matching method. The problem is challenging due to the low resolution of videos, vary of illumination, positional variances and the possible appearance of carried objects at exclusive viewpoints. Another problem is when wearing the same clothes may be detected as they are the same person. Therefore, these kinds of person re-identification systems get low accuracy results. Hence, we focus our research on using comprehensive Convolutional Neural Network (CNN) for person re-identification and tracking in public areas with the aim of the full automation person re-identification system using multiple cameras. In our method, we grab the frames from the video sequence using an accurate multi-target tracker. Then, each frame is processed by a Siamese network architecture developed by using Caffe Framework to generate a feature vector consist of 1024 values representing the person’s physical appearance. This architecture consists of two sub-networks with identical weights with. When the network gets two inputs, the sub-networks map those inputs to a pair of feature vectors. Then it compares two output vectors using Euclidean distance. During training, the Siamese network is shown similar and dissimilar input pairs. Features are extracted from the frame by frame using a CNN in our system. Here, we crop the person using the Python image library module as background reduction for accurate the results. Then, we combine information from all time-steps to give a comprehensive appearance feature for the outright sequence. The System achieves 55% accuracy using our Siamese network architecture.Item Pronunciation Training System to Enhance Sinhala and Tamil Language Skills(Uva Wellassa University of Sri Lanka, 2016) Carnage, G.R.M.D.; Thambawita, D.R.V.L.B.This is a pronunciation training system to enhance Sinhala & Tamil language skills. Most commercial pronunciation systems are developed in English and those systems are limited to people with English literacy. Language is never a simple issue of communication, in contemporary social and political practice everywhere, language goes much beyond its basic utilitarian purposes. In this sense, Sri Lanka has no exception. By now, Sri Lanka has ended an immensely destructive military conflict that had much to do with a crisis of identity linked as much to language as ethnicity and contested notions of binary-nationalisms and competitive interpretations of history. As a solution for language problem between Tamils and Sinhalese, this research idea comes out. This system aimed both Sinhala and Tamil languages and is developed pronunciation training tests for both languages. This system is the only pronunciation tool develop for both Sinhala and Tamil languages. This system has been tested in both Sinhala and Tamil. Firstly system pronounce the word and user can listen to the word and pronounce the word. Then system compares both audio files and gives a score. The main objective of this pronunciation training system is comparing two audio files and giving a score as a percentage. For that Fast Fourier transformation has been used. Fast Fourier transform converts time domain waveform samples into the frequency domain spectrum samples. This system is a very user friendly system for the user. Users can select language and the words according to the preference and preview the pronouncing word before pronouncing it. As well as graphically illustrate the frequency with the time of both original and user's sound. Therefore, users can get a rough idea about the original audio and pronounced audio. Keywords: Pronunciation, Sinhala, FourierItem Real Time Traffic Light Time Analyzer(Uva Wellassa University of Sri Lanka, 2016) Indika, I. K. K.; Thambawita, D.R.V.L.B.In urban arrears of all over the world, traffic lights are used to manage the traffic. However, congestion becomes worst and needed to increase the performance of traffic lights by focusing on increasing the capacity of transactions and minimizing the waiting time. By this research, a new fuzzy rule set is proposed for the cross and three-way (T-junction) junctions considering a number of vehicles on a road, time to complete a cycle, waiting time of the each road and the traffic condition of the road. The image processing technique is used to get dynamic data. The fuzzy logic system has been implemented to get decisions dynamically for the cycle time, green color light time and time to cross vehicle by analysing the waiting times using fuzzy outputs and other dependent data. MATLAB is used for fuzzy logic and image processing to generate unique output for each traffic condition. The performance of the new system is evaluated by comparing the waiting times in between current cycles and fuzzy cycles. The results obtained shows that the system has almost higher performance (Between 13% and 47%) and it is increased when congestion is low. Use of dynamic time cycles according to vehicle velocity causes to increase the performance of it. It was identified that performance is higher when using short cycle times for less congestion while medium cycle time is preferred for high congestion. The output results show that fuzzy system is better than the existing system because its waiting time near junctions is less than fixed cycle method. Keywords: Traffic congestion, Fuzzy logic, Delay estimation, Real timeItem Stock handling and analyzing system for multi-purpose Co-operative society, Eheliyagoda(Uva Wellassa University of Sri Lanka, 2015) Senadheera, S.P.T.P.; Thambawita, D.R.V.L.B.The multi-purpose co-operative society is one of the largest business bodies around the area which has number of regional shops (around 40) throughout the division which are controlled by the head office situated at Eheliyagoda. This multi-purpose co-operative society offers large verities of consuming goods to fulfill the needs of the consumers around the area. In order to cope with this particular customer demand, a specific stock handling mechanism is needed. This stock handling mechanism should be run by the main store complex as the base under the co-operative supervision of a store manager and the head office authorities while regional shops playing an important role. With the existing manual stock handling procedure, it is difficult to maintain the stock details and to communicate between the store complex, head office and the regional shops. As an example, if a regional shop manager wants to tally the consumer demand information, he/she needs to do it manually and needs to come to the main store complex in order to handover the consumer demand information and then the store keeper has to carry his/her reports to the head office. In this context proper application is needed to integrate the regional shop manager, store keeper and the head office authorities in stock handling procedure. After receiving the customer demand report, the head office authorities need to look for the suppliers from whom that particular goods can be bought for a fair price. This is also a time wasting and exhausting task. In this context, they need an effective mechanism to call prices for the required goods hence they can decide the optimal supplier to be reached which means this system can help the co-operative society to build a strong links with external suppliers.. Methodology In order to cope with above mentioned short comings, a proposed system has come up with two solutions. The first one is the desktop software application which helps the internal parties (regional shop managers, store manager, head office authorities) to integrate each other for effective stock handling procedure. The other solution is the web application which helps to call prices from the external suppliers and ultimate to build some strong links with external parties. Data collection, Result and Discussion The data collection was done by an advanced discussion and a smooth study. nThe advanced discussion covered the entire process flow of multi-purpose co-operative society, Eheliyagoda. The authorities have pointed out the weak links/vacuums which are generally can be seen their process flow. That was really a complex one comparing to a general scope of a final year project of a bachelor’s degree. The fact that has been pointed out by them was the requirement of an effective procedure of stock handling. There are specific forms can be seen in order to use in their stock handling process flow. These forms helped greatly to draft the Skelton of the system and carry it on. According to the drafted system, four roles can be understood. They are; I. II. III. IV. Regional shop managers. Store keeper. Head office authorities. External suppliers. The proposed system comprises with a desktop application to integrate regional shop manager, store keeper and the head office and the web application integrate the firma and the external suppliers in order to call prices. Java, PHP, CSS3, HTML5, javascript have been used to develop the system. Apart from that, NetBeans was the Integrated Development Environment while notepad++ has been used as the text editor.Item Toxic Comment Classification Using Machine Learning(Uva Wellassa University of Sri Lanka, 2019-02) Pramodya, L.A.S.; Rathnayaka, R.M.G.U.; Lahiru, K.K.S.; Thambawita, D.R.V.L.B.Comment classification models are available today for “flagging” the comments. However, determining whether or not a comment should be “flagged” is difficult and time-consuming. Another major problem is the lack of sufficient data for training the model, and there are some issues with the available datasets because those are annotated by the human raters and those annotations are dependent on their personal beliefs. Lack of multi-label comment classification model causes for issues of abusive behavior. This paper presents models for multi-label text classification for identifying the different level of toxicity within a comment. In this paper, we use Wikipedia comments which have been labeled by human raters for toxic behavior provided by Kaggle. Comments have been categorized into six categories as toxic, severe-toxic, obscene, threat, insult, and identityhate. The dataset contains 159572 comments. For data analyzing we use python seaborn library and python matploitlib library. It is understood that the dataset is highly skewed. Most of the comments do not belong to any of the six categories. Researchers used undersampling for majority class to correct the bias in the original dataset. We tested three models: a feed-forward neural network with Keras and word embedding, a Naive Bayes model with Scikit-Learn, and a LightGBM with 4-fold cross-validation. For the neural network, it took 3.5 hours to be trained on Nvidia GeForce 840M which is having 384 CUDA cores, Naive Bayes model with Scikit-Learn took 3 hours where LightGBM with k-fold took 4 hours. Researchersran 100 epochs from each model. At the end of 100 epoch, the neural network gave 0.9930 of validation accuracy and loss was just 0.2714, Naive Bayes model with Scikit-Learn gave 0.9556 validation accuracy and loss was 0.4121 where LightGBM with k-fold accuracy was 0.9000 and validation loss was 0.4263. The neural network gave the best accuracy at the end of the 100th epoch.Item Virtual Laboratory System for AIL Physics Units of Electricity(Uva Wellassa University of Sri Lanka, 2016) Maheshika, H.W.D.; Thambawita, D.R.V.L.B.Practical experiment, an important feature of science education receives little attention in the science curriculum for several reasons. In science subjects, laboratories perform an important role in students learning enhancement, particularly in physics education. Students take more interest in learning by performing and observing the experiments in a physics laboratory which improve and consolidate their learning. Therefore, we want to recognize the need for field and laboratory exercises in school science programs as they help to promote a science, technology and innovation culture in the country. However, due to financial problems and other deficient facilities in laboratories, most of the schools cannot establish a complete physics laboratory. However, the lack of laboratory resources in schools appears to be the main reason in rural areas. Also, lack of focus on practical work, teacher shortages and variable quality of teachers were equally affected to science education at the GCE AIL. But the syllabus of the GCE AIL science subjects is competency- based, student- centered and activity- oriented and aimed to develop personal, interpersonal and higher order thinking skills of the target student group. In this paper, we describe the design and implementation of a Virtual Laboratory, which can be proven as an important educational tool that deals with the lack of practical experience in education. It also fulfills the deficiency of physical equipment which appears in the real physics laboratory environment. The aim of VPL is the simulation of physics experiments with 3D interaction in an easy way and access to educational and research material is facilitated for both students and teachers as well. In order to evaluate the VPL, we invited students of different schools in rural areas. This virtual lab consisting of tools and simulators can be effectively replaced the use of the corresponding necessary equipment and will provide equivalent services to support students. Virtual labs offer some advantages in education and can supplement physical labs. Keywords: Experiment, Virtual Laboratory, 3D InteractionItem Vision Based Autonomous Micro Areal Vehicle with Graphic Processing Unit Acceleration(Uva Wellassa University of Sri Lanka, 2018) Basnayake, D.C.; Rathnamalala, V.N.M.; Manchanayaka, M.A.D.E.; Thambawita, D.R.V.L.B.; Madusanka, D.G.P.N.Autonomous Micro Aerial Vehicles (MAV) are available today to process and navigate within indoor environments. However, lack of processing power of the drone generates unreliable, inaccurate and inefficient navigation results. This paper presents a model for a vision based autonomous hexa-copter for navigating within an indoor environment with the use of GPU acceleration. The MAV is embedded with Jetson TK1 board with GPU computing power, MS Kinect, Arduino Flight controller. Jetson TK1 on development kit gets the outside data using Kinect sensor and build the map itself by Jetson TK1 board with the help of GPU computing. Then drone can navigate without a remote access and it uses the flight controller to control MAV. In addition to that, Real Time Appearance Based Mapping (RTAB) was used to build the 3D map of an indoor environment. RTAB-map is used for the SLAM (Simulation Localization and Mapping) approach. GPU parallel computing power of the Jetson board was used for optimizing SLAM algorithm. MAV estimates position over time using visual odometry. Adaptive Monte Carlo Localization algorithm used for estimate robot position. A Robot Operating System (ROS) uses the Rviz (ROS visualization) framework to visualize all the data. Rviz visualizes the 3D module of MAV and 3D map. ROS uses Rviz framework as a virtual environment and it passes the command to flight controller through serial communication while navigating through this virtual environment. MAV uses the navigation stack for the autonomous navigation. The main feature of this research is MAV process the data itself. The process includes the 3D mapping, real time state estimation, obstacle detection and navigation planning. RTAB-map builds the environment 3D map and navigation stack use for MAV navigation. Although the drone navigates real time itself, embedding all the component within the drone make its weight greater which make the power consumption. issues.