Browsing by Author "Pathirana, P"

Now showing 1 - 5 of 5

item: Conference-Extended-Abstract
Current status of relative motion detection and segmentation of moving objects for navigation using optical flow
(2007) Fernando, WSP; Udawatte, L; Pathirana, P
Motion Control is one of the most important research topics in computer vision. It is the base for many other problems such as visual tracking, structure from motion. A major interests of motion analysis is to estimate 3D motion.
item: Article-Full-text
Customer gaze estimation in retail using deep learning
(IEEE, 2022) Senarath, S; Pathirana, P; Meedeniya, D; Jayarathn, S
At present, intelligent computing applications are widely used in different domains, including retail stores. The analysis of customer behaviour has become crucial for the bene t of both customers and retailers. In this regard, the concept of remote gaze estimation using deep learning has shown promising results in analyzing customer behaviour in retail due to its scalability, robustness, low cost, and uninterrupted nature. This study presents a three-stage, three-attention-based deep convolutional neural network for remote gaze estimation in retail using image data. In the rst stage, we design a mechanism to estimate the 3D gaze of the subject using image data and monocular depth estimation. The second stage presents a novel three-attention mechanism to estimate the gaze in the wild from eld-of-view, depth range, and object channel attentions. The third stage generates the gaze saliency heatmap from the output attention map of the second stage. We train and evaluate the proposed model using benchmark GOO-Real dataset and compare results with baseline models. Further, we adapt our model to real-retail environments by introducing a novel Retail Gaze dataset. Extensive experiments demonstrate that our approach signi cantly improves remote gaze target estimation performance on GOO-Real and Retail Gaze datasets.
item: Article-Full-text
Eye gaze estimation: A survey on deep learning-based approaches
(Elsevier, 2022) Pathirana, P; Senarath, S; Meedeniya, D; Jayarathna, S
Human gaze estimation plays a major role in many applications in human-computer interaction and computer vision by identifying the users’ point-of-interest. The revolutionary developments of deep learning have captured significant attention in the gaze estimation literature. Gaze estimation techniques have progressed from single-user constrained environments to multiuser unconstrained environments with the applicability of deep learning techniques in complex unconstrained environments with extensive variations. This paper presents a comprehensive survey of the single-user and multi-user gaze estimation approaches with deep learning. The state-of-the-art approaches are analyzed based on deep learning model architectures, coordinate systems, environmental constraints, datasets and performance evaluation metrics.Akey outcome from this survey realizes the limitations, challenges, and future directions of multi-user gaze estimation techniques. Furthermore, this paper serves as a reference point and a guideline for future multi-user gaze estimation research.
item: Thesis-Abstract
Real-time detection and tracking of vehicles with lane detection
(3/23/2012) Fernando, WSP; Udawatta, L; Pathirana, P
In this research, a computer vision based procedure for navigating an autonomous vehicle safely in a sub-urban road under an unstructured environment was described. This was analyzed in two main areas. Namely; an on road object detection method, where we are only concerned of detecting cars, and a novel method in detecting road lane boundaries. For the detection of vehicles (cars) from an on-road image sequence taken by a monocular video capturing device in real time and an algorithm of multi resolution technique based on Haar basis functions were used for the wavelet transform, where a combination of classification was carried out with the multilayer feed forward neural network. The classification is done in a reduced dimensional space, where Principle Component Analysis (PCA) dimensional reduction technique has been applied to make the classification process much more efficient. Then, the other approach used is based on boosting which also yields very good detection rates. In general, boosting is one of the most important developments in classification methodology. It works by sequentially applying a classification algorithm to reweighed versions of the training data, followed by taking a weighted majority vote of the sequence of classifiers thus produced. For this work, a strong classifier was trained by the discrete adaboost algorithm and its variants. In this thesis, a novel algorithm for detection of lane boundaries was presented. Initially, the method fits the CIE L*a*b* transformed road chromaticity values (that is a* and b* values) to a bi-variate Gaussian model followed by the classification of road area based on Mahalanobis distance. Then, the classified road area acts as an arbitrary shaped region or a mask in order to extract blobs resulting from the filtered image by a two dimensional Gabor filter. This is considered as the first visual cue. Another visual cue of images was employed by an entropy image. Moreover, the results from color based visual cue and visual cue based on entropy were integrated following an outlier removing process. Finally, the correct road lane points are fitted with Bezier splines which act as control points that can form arbitrary shapes. The algorithm was implemented and experiments were carried out on sub-urban roads.
item: Conference-Full-text
A rule based approach for hemorrhage detection in digital fundus photographs
(Faculty of Information Technology, University of Moratuwa., 2021-12) Munasingha, SC; Pathirana, P; Priyankara, KK; Upasena, RG; Ikeda, A; Ganegoda, GU; Mahadewa, KT
Hemorrhages are one of the earliest signs of Diabetic Retinopathy, hence accurate detection of hemorrhages is crucial in an automated DR detection system. In this paper, a novel and robust rule based methodology for automated detection of hemorrhages is proposed. We present an ensemble technique for hemorrhage classification by incorporating size-based classification, color-statistic-based classification, and shape-based classification along with a novel dual step filtering approach for candidate detection. Finally, we present an experimental study carried out on DIARETDB database using the proposed method to detect and segment hemorrhages in retinal images.