Browsing by Author "Rodrigo, R"

Now showing 1 - 20 of 44

item: Conference-Abstract
Action recognition by single stream convolutional neural networks : an approach using combined motion and static information
Ramasinghe, S; Rodrigo, R
We investigate the problem of automatic action recognition and classification of videos. In this paper, we present a convolutional neural network architecture, which takes both motion and static information as inputs in a single stream. We show that the network is able to treat motion and static information as different feature maps and extract features off them, although stacked together. We trained and tested our network on Youtube dataset. Our network is able to surpass state-of-the-art hand-engineered feature methods. Furthermore, we also studied and compared the effect of providing static information to the network, in the task of action recognition. Our results justify the use of optic flows as the raw information of motion and also show the importance of static information, in the context of action recognition.
item: Conference-Abstract
Action recognition using a spatio-temporal model in dynamic scenes
Chathuramali, KGM; Rodrigo, R
Action recognition in a video plays an important role in computer vision and finds many applications in areas such as surveillance, sports, and elderly monitoring. Existing methods mostly rely on stationary backgrounds. Action recognition in dynamic backgrounds typically requires standard preprocessing steps such as motion compensation, background modeling, moving object detection and object recognition. The errors of the motion compensation step and background modelling increase the mis-detections. Therefore action recognition in dynamic background is challenging. In this paper, we use a combination of pose characterized by a silhouette and optic flows synthesized into a histogram. This enables us to classify the movement of the actor versus movement of the background. We use four background models to extract the silhouette from the frame. We use SVM to recognize actions, according to several evaluation protocols. We perform several experiments and compare over a diverse set of challenging videos, including the new Change Detection Challenge Dataset. Our results perform better than existing methods.
item: Thesis-Full-text
Activity recognition combined with scene context and action sequence
Ramasinghe, SC; Rodrigo, R
In this study, we investigate the problem of automatic action recognition and classification of videos. First, we present a convolutional neural network architecture, which takes both motion and static information as inputs in a single stream. We show the network is able to treat motion and static information as different feature maps and extract features off them, even though stacked together. By our results, we justify the use of optic flows as the raw information of motion. We demonstrate that our network is able to surpass state-of-the-art hand-engineered feature methods. Furthermore, the effect of providing static information to the network, in the task of action recognition, is also studied and compared here. Then, a novel pipeline is proposed, in order to recognize complex actions. A complex activity is a temporal composition of subevents, and a sub-event typically consists of several low level micro-actions, such as body movement, done by different actors. Extracting these micro actions explicitly is beneficial for complex activity recognition due to actor selectivity, higher discriminative power, and motion clutter suppression. Moreover, considering both static and motion features is vital for activity recognition. However, how to control the contribution from each feature domain optimally still remains uninvestigated. In this work, we extract motion features in micro level, preserving the actor identity, to later obtain a high-level motion descriptor using a probabilistic model. Furthermore, we propose two novel schemas for combining static and motion features: Cholesky transformation based and entropy based. The former allows to control the contribution ratio precisely, while the latter uses the optimal ratio mathematically. The ratio given by an entropy based method matches well with the experimental values obtained by a Choleksy transformation based method. This analysis also provides the ability to characterize a dataset, according to its richness in motion information. Finally, we study the effectiveness of modeling the temporal evolution of sub-event using an LSTM network. Experimental results demonstrate that the proposed technique outperforms state- of-the-art, when tested against two popular datasets.
item: Conference-Abstract
Appearance based tracking with background subtraction
Jayamanne, DJ; Samarawickrama, J; Rodrigo, R
Grouping the detected feature points traditionally requires the storage of long corner tracks. The traditional method does not permit to arrive at a decision to cluster the feature points based on a frame by frame basis. This paper presents a method to group the feature points directly into objects using the most recent 20 frames. The detected corner features are validated and clustered based on two approaches. When objects move in isolation, an EM algorithm is used to cluster and every object is detected and tracked. When objects move under partial occlusion, the corner features are clustered based on an agglomerative hierarchical clustering approach. A probabilistic framework has also been applied to determine the object level membership of the candidate corner features. A novel foreground estimation algorithm with an accuracy of 98% based on color information, background subtraction result and detected corner features is also presented.
item: Conference-Abstract
Automatic number plate recognition in low quality videos
Ajanthan, T; Kamalaruban, P; Rodrigo, R
Typical Automatic Number Plate Recognition (ANPR) system uses high resolution cameras to acquire good quality images of the vehicles passing through. In these images, license plates are localized, characters are segmented, and recognized to determine the identity of the vehicles. However, the steps in this workflow will fail to produce expected results in low resolution images and in a less constrained environment. Thus in this work, several improvements are made to this ANPR workflow by incorporating intelligent heuristics, image processing techniques and domain knowledge to build an ANPR system that is capable of identifying vehicles even in low resolution video frames. Main advantages of our system are that it is able to operate in real-time, does not rely on special hardware, and not constrained by environmental conditions. Low quality surveillance video data acquired from a toll system is used to evaluate the performance of our system. We were able to obtain more than 90% plate level recognition accuracy. The experiments with this dataset have shown that the system is robust to variations in illumination, view point, and scale.
item: Conference-Abstract
Automatic number plate recognition in low quality videos
(2014-06-19) Ajanthan, T; Kamalaruban, P; Rodrigo, R
Typical Automatic Number Plate Recognition (ANPR) system uses high resolution cameras to acquire good quality images of the vehicles passing through. In these images, license plates are localized, characters are segmented, and recognized to determine the identity of the vehicles. However, the steps in this workflow will fail to produce expected results in low resolution images and in a less constrained environment. Thus in this work, several improvements are made to this ANPR workflow by incorporating intelligent heuristics, image processing techniques and domain knowledge to build an ANPR system that is capable of identifying vehicles even in low resolution video frames. Main advantages of our system are that it is able to operate in real-time, does not rely on special hardware, and not constrained by environmental conditions. Low quality surveillance video data acquired from a toll system is used to evaluate the performance of our system. We were able to obtain more than 90% plate level recognition accuracy. The experiments with this dataset have shown that the system is robust to variations in illumination, view point, and scale.
item: Conference-Full-text
Beyond the traditional telecom promotions, with the escalating customer Base
(The Engineering Research Unit, University of Moratuwa, 2013-02) Jayawardhana, MKPR; Kumara, PAKM; Perera, TDK; Paranawithana, WDAI; Anjitha, T; Weerawarana, SM; Rodrigo, R
Through the past years, traditional telecom customer base has escalated with wide usage of mobiles. But the services and promotions offered by the service providers has been limited to a specific frame, without extracting useful information from the data they gather with each call. Specifically in mobile user segment, there is lot of data available for the service provider other than the credit balance to consider in giving away service and promotions. Among many reasons, a barrier for this is the technical feasibility. With the growing customer base and elastic behavior of network usage with the time and seasons, proposing an effective way of available data analyzing is challenging. It becomes more challenging when the results are required in near real time for the promotion to be of use. With this research studyt we discuss and compare the available technologies to be used, propose a system with the most appropriate technologies and express the results gained with an implemented prototype ‘Kanthaka'.
item: Conference-Full-text
Building sand castles within iaas - based cloud instances
(The Engineering Research Unit, University of Moratuwa, 2013-02) Sharma, K; Carnage, C; Rodrigo, R
Enterprises continually seek innovative approaches to reduce operational computing costs while extracting the maximum utility from their resources. Cloud Computing technology play a major role in helping organizations to reduce the operational cost and it is a model for enabling convenient, on-demand network access to a shared pool of configurable computing resources. Cloud consumers are allowed to upload and execute their code inside Cloud Instances(CI) to perform different tasks. At the same time they need to run applications which they cannot trust completely. A compelling approach is needed to mitigate the security risk of executing untrusted applications that could potentially corrupt the resources available for CIs. This paper proposes a confined execution environment, which provides security and protection for CIs running untrusted applications. Cloud consumers can enable the proposed confined execution environment as and when required.
item: Conference-Full-text
Characterization of clay deposits in Nachchcaduwa area ceramic
(The Engineering Research Unit, University of Moratuwa, 2013-02) Wanasinghe, DD; Adikary, SU; Rodrigo, R
In this research clay deposits located in the Nachchaduwa area were investigated to identify suitable ceramic fabrication techniques and products. Deposits located in this area are known to be rich in Kaolinite and Montmorillonite (MMT) and other type of phyrophyllitc clays. These are known as "Red Clay and mined to fabricate traditional ceramic ware by traditional techniques. The specimens were collected from tanks located in this area and subjected to Fourier Transform Infrared Spectroscopy (FTIR), X-ray Diffraction (XRD), Differential Thermal Analysis (DTA) and Thermo Gravimetric Analysis (TGA), after purifying them and removing organic compounds. Combination of these techniques with the chemical analysis on selected specimens was employed to accurate identification of the clay specimens. The results were then compared with each other and published literature for the identification. Results showed the presence of MMT, kaolinite, quartz and other type of clay minerals in small quantities; furthermore specimens subjected to chemical analysis revealed that they contain more than 50% quartz by weight. The purified clay specimens subjected to a Dejlocculant Demand Test, which determines the optimum amount of defocculant needed to prepare a casting slip, but the test showed that these claysare not suitable to prepare a casting slip in fabricating ceramic ware.
item: Article-Full-text
Combined static and motion features for deep-networks-based activity recognition in videos
(IEEE, 2019) Ramasinghe, S; Rajasegaran, J; Jayasundara, V; Ranasinghe, K; Rodrigo, R; Pasqual, AA
Activity recognition in videos in a deep-learning setting—or otherwise—uses both static and pre-computed motion components. The method of combining the two components, whilst keeping the burden on the deep network less, still remains uninvestigated. Moreover, it is not clear what the level of contribution of individual components is, and how to control the contribution. In this work, we use a combination of CNNgenerated static features and motion features in the form of motion tubes. We propose three schemas for combining static and motion components: based on a variance ratio, principal components, and Cholesky decomposition. The Cholesky decomposition based method allows the control of contributions. The ratio given by variance analysis of static and motion features match well with the experimental optimal ratio used in the Cholesky decomposition based method. The resulting activity recognition system is better or on par with existing state-of-theart when tested with three popular datasets. The findings also enable us to characterize a dataset with respect to its richness in motion information.
item: Conference-Extended-Abstract
Computationally efficient implementation of video rectification in FPGA fo stereo vision applications
(2010) Maldeniya, BS; Navvarathna, UDU; Wijayasekara., KU; Wijegoonasekara, WTUS; Rodrigo, R
In order to obtain depth perception in computer vision, one needs to process pairs of stereo images. This process is computationally challenging to be carried out in real-time, because it requires the search for matches between objects in both images. Such process is significantly simplified if the images are reflected. Stereo image reflection involves a matrix transformation which when done in software will not produce real-time results. But in stereo vision applications this features is very demanding. On the other hand, applying those transformations to the video frames is very restricted by real-time constraints. Therefore, the video streaming and matrix transformation are not usually implemented in the same system. Our product is a stereo camera pair which produces a rectified real time image output with a resolution of 320x240 at a frame rate of 15FPS and delivers then via 100-Ethernet interface. We use an Spartan 3E FPGA for real-time processing within we implement an image rectification algorithm.
item: Conference-Full-text
Control system for quadrotor uav
(The Engineering Research Unit, University of Moratuwa, 2013-02) Sampath, BG; Wijesiri, NRAAR; Pitahawatte, JMLMGB; Dassanayake, VPC; Rodrigo, R
This paper presents the design, analysis and testing of a control system for a quadrotor. The research is done together with the were focused on the maneuverability of the quadrotor hence the mechanical design design of the controlling algorithm. Constraints which occur due to using pre-built quadrotors. eliminated by using this methodology. This enables more aggressive and aerobatic fly mg compared to other systems designed with off-the-shelf quadrotors.
item: Conference-Full-text
Corrosion behavior of steel in different atmospheric conditions
(The Engineering Research Unit, University of Moratuwa, 2013-02) Adikari, AAMT; Munasinghe, RGN De S; Jayatiieke, S; Rodrigo, R
Corrosion of metals makes a large impact on the remedial actions to prevent structures, corrosion the corrosion that economy of a country. Therefore, it is important machinery and vehicles from corrosion. Among the various types of occurs in the atmosphere is known as atmospheric corrosion and it accounts for more failures than other types oj corrosion. To take preventive actions against atmospheric corrosion of metals, it is essential to study the corrosivity of the atmosphere by analyzing factors that influence it. The corrosivity of the atmosphere mostly depends on several atmospheric variables, such as relative humidity, temperature, rainfall, chloride deposition rate, pollutant gases like nitrogen and sulfur oxides. In order to study the severity of the atmosphere which promotes the used structural materials mild steel, stainless steel 304 and 316 conditions. Two corrosion test panels to take metal corrosion three types of commonly were exposed in two different atmospheric placed in two geographical locations and loss of weight due to corrosion was continuously measured in all three types of metals. The atmospheric variables in the two locations were also continuously recorded. Finally, these data were fitted with the power model in order to predict the rate of corrosion under particular atmospheric conditions. With the predicted corrosion rate under a particular atmospheric condition it is possible to take necessary> preventive measures during the design or in the service of metallic structures, machinery and vehicles etc. The broad aim of this research work is to collect adequate data to develop a corrosion model to predict the corrosion rate in any atmospheric environment in Sri Lanka using measured atmospheric variables and thereby establish a 'corrosion map'for Sri Lanka.
item: Conference-Full-text
Design and Optimization of a MEMS Based Piezoresistive Pressure Sensor for Hash Flood Level Measurement.
(The Engineering Research Unit, University of Moratuwa, 2013-02) Priyadarshana, TGP; Wijethunge, HMDP; Jayasekara, BCCP; Amarasinghe, YWR; Rodrigo, R
This paper is focused on designing and optimization of a Micro Electro Mechanical System (MEMS) based piezoresistive .t ype pressure 10 enhance the sensitivity by optimally utilizing the maximum available Silicon wafer cell volume of 3mmX3mmX400pm and a fixed thickness of 10pm due to fabrication constrains. Here an analysis is carried out by varying the dimensions of two different diaphragm geometries, namely conventional square type diaphragm and cross-sectional beam type diaphragm. The analysis is done using finite element method (FEM) technique in ANSYS software and by comparing the results, the better diaphr agm type is chosen for the required pressure range oj the application for flash fiood level The results show that for some pressure ranges the cross-sectional beam type diaphragm delivers a much better sensitivity than the conventional square type diaphragm.
item: Conference-Extended-Abstract
Design and testing a portable negative preasure wound therapy (NPWT) device
(The Engineering Research Unit, University of Moratuwa, 2013-02) Welgama, WPD; Gray, HA; Amarasinghe, YWR; Wijeyaratne, SM; Sugathapala, AGT; Rodrigo, R
The purpose of this research was to develop a clinically validated portable NPWT device with low power consumption features. A new NPWT device having dimension of 30 cmy'20 cm x / 5cm (Length x Width x Hight) was clinically tested to evaluate its performance. The device is capable of working continuously for more than 24 hours under pressure range from -50 mmHg to -250 mmHg with ± 3 mmHg accuracy’. After several clinical tests, it could be confirmed that the wound healing rate has been increased significantly with the aid of developed NPWT device.
item: Conference-Full-text
Distributed intelligence as anlnformation retrievalsystem
(The Engineering Research Unit, University of Moratuwa, 2013-02) Jayathilake, DSP; Herath, HMST; Wimalasooriya, RDDP; Karunarathne, SDJ; Weerawardhane, S; Rodrigo, R
such as Geographic Information Systems (CIS), because accurate information of geographic data a spat,ally distributed and vast in volume. The challenge is to improve the process of information retrieval, which can be applied to varying contexts such as GIS, going beyond the limits imposed by the knowledge base centric approaches. Social networks possess a vast amount of unstructured knowledge in its user base, which can be use in a awing interesting patterns, if analyzed in a systematic approach. In this paper, we emphasize on an approach for information search and retrieval, bv combining the social network of human intelligence and knowledge bases of information. Users of the system can present natural language queries, and questions be directed to relevant experts within the social network. Distributed information retrieval and information fusion is applied to optimize the process oj single answer generation. The feasibility of our proposed architecture is investigated by implementing a QA(Question Answering) system on a social network as a proof of concept. The Question Answering System is based on geogi aphic queries which will be directed to relevant users within the network to gather answers. Varying methods of information fusion and Natural Language Processing have been researched and discussed to discover the suitable approaches for generation. With the wide acceptance of the social networking paradigm and sharing of information, this approach be extended to other domains as well.
item: Article-Full-text
Diverse single image generation with controllable global structure
(Elsevier, 2023-04) Mahendren, S; Edussooriya, CUS; Rodrigo, R
Image generation from a single image using generative adversarial networks is quite interesting due to the realism of generated images. However, recent approaches need improvement for such realistic and diverse image generation, when the global context of the image is important such as in face, animal, and architectural image generation. This is mainly due to the use of fewer convolutional layers for capturing the patch statistics and, thereby, not being able to capture global statistics well. The challenge, then, is to preserve the global structure, while retaining the diversity and quality of image generation. We solve this problem by using attention blocks at selected scales and feeding a random Gaussian blurred image to the discriminator for training. We use adversarial feedback to make the quality of the generation better. Our results are visually better than the state-of-the-art, particularly, in generating images that require global context. The diversity of our image generation, measured using the average standard deviation of pixels, is also better.
item: Conference-Full-text
Document analysis based automatic concept map generation for enterprises
(The Engineering Research Unit, University of Moratuwa, 2013-02) Herath, HMTC; Fernando, KNJ; Karannagoda, EL; Karunarathne, WMID; De Silva, NHN; Perera, AS; Rodrigo, R
Ever growing knowledge bases of enterprises present the demanding challenge of proper organization of information that would enable fast retrieval of related and intended information. Document repositories of enterprises consist of large collections of documents of varying size, format and writing styles. This diversified and unstructured nature of documents restrict the possibilities of developing uniform techniques for extracting important concepts and relationships for summarization, structured representation and fast retrieval. The documented textual content is used as the input for the construction of this concept map. Here a rule based approach is used to extract concepts and relationships among them. Sentence level breakdown enables these ndes to identify’ those concepts and relationships. These rules are based on elements in a phase structure tree of a sentence. For improving accuracy and the relevance of the extracted concepts and relationships, the special features such as titles, bold and upper case texts are used. This paper discusses how to overcome these challenges by utilizing high level natural language processing techniques, document preprocessing techniques and developing easily understandable and extractable compact representation of concept maps. Each document in the repository is converted to a concept map representation to capture concepts and relationships among concepts described in the said document. This organization would represent a summary> of the document. These individual concept maps are utilized to generate concept maps that represent sections of the repository> or the entire document repository’. This paper discusses how the statistical techniques used to calculate certain metrics which facilitate certain requirements of the solution. Principle component analysis is used in ranking the documents by importance. The concept map is visualized using force directed type graphs which represent concepts by nodes and relationships by edges.
item: Conference-Full-text
Effect of cryogenic cooling on machining performance on hard to cut metals - a literature review
(The Engineering Research Unit, University of Moratuwa, 2013-02) Senevirathne, SWMAI; Fernando, MARV; Rodrigo, R
This paper presents a literature review done on Cryogenic cooling, Liquid Nitrogen Coot,ng(LNC) and •Chilled Air and Minimum Quantity Lubrication (CAMQL) cooling methods used ,n targeting to recognise belter cooling methods for the Sri Lankan d,e and mould making sector (SLDMMS). The state of the art in cryogenic machining was searched and reviewed first, and then its benefits, limitations and applications were studied. Alternative cooling methods for conventional emulsion cooling were seai ched. The benefits from cryogenic cooling and other alternative methods were analysed and compared. LNC and CAMQL cooling were chosen as subjects for further comparative study. A substantial amount of literature was found on the effect of these cooling methods on hard-to-cut materials such as Inconel, Titanium alloys etc., but very few or no studies had been carried out on materials used in the Sri Lankan die and mould making sector A survey on materials used in SLDMMS is recommended to identify the most commonly used material types. An economic feasibility study of cryogenic cooling and chilled air cooling is recommended.
item: Article-Full-text
End-to-end data-dependent routing in multi-path neural networks
(Springer, 2025) Tissera, D; Wijesinghe, R; Vithanage, K; Xavier, A; Fernando, S; Rodrigo, R
Neural networks are known to give better performance with increased depth due to their ability to learn more abstract features. Although the deepening of networks has been well established, there is still room for efficient feature extraction within a layer, which would reduce the need for mere parameter increment. The conventional widening of networks by having more filters in each layer introduces a quadratic increment of parameters. Having multiple parallel convolutional/dense operations in each layer solves this problem, but without any context-dependent allocation of input among these operations: the parallel computations tend to learn similar features making the widening process less effective. Therefore, we propose the use of multipath neural networks with data-dependent resource allocation from parallel computations within layers, which also lets an input be routed end-to-end through these parallel paths. To do this, we first introduce a crossprediction based algorithm between parallel tensors of subsequent layers. Second, we further reduce the routing overhead by introducing feature-dependent cross-connections between parallel tensors of successive layers. Using image recognition tasks, we show that our multi-path networks show superior performance to existing widening and adaptive feature extraction, even ensembles, and deeper networks at similar complexity.