Pedestrian Detection and Tracking Using Depth Cues
In this project, a robust computer vision approach to detect and track pedestrians in unconstrained crowded scenes is proposed. Pedestrian detection is performed via a 3D clustering process within a 3D volume of interest (VOI). Projected top view features on to the groundplane are used for detecting and tracking completely occluded objects. The approach is evaluated using both indoor and outdoor sequences captured using a variety of different camera placements and orientations that fetch significant challenges in terms of the number of pedestrians present, their interactions and scene lighting conditions. Results point to the extremely accurate performance of the proposed approach in all cases.
Feature Fusion for Robust Face Recognition
Gabor features have been known to be effective for face recognition. However, only a few approaches utilize phase feature and they usually perform worse than those using magnitude feature. To investigate the potential of Gabor phase and its fusion with magnitude for face recognition, a new system for face recognition is developed based on the Gabor phase and magnitude with the Local binary patterns (LBP). Two features are fussed with nave fashion so that it can enhance the detection accuracy. The evaluation has been done on a set of face datasets and the fusion approach outperforms most of the state-of-the-art approaches.
Sequential Features for Face Recognition
This project aims to develop a robust face recognition system using sequential features. The developed technique extracts Discrete Wavelet Transformed (DWT) features from the raw image in the initial stage and the Principal component features will be extracted for the DWT features. These sequential features will have high discriminative capability and results in good detection accuracy. The developed algorithm is tested using ORL and Yale Datasets and the method shows higher detection rate as compared to state-of-the-art.
Kernel Based Automatic Traffic Sign Detection and Recognition Using SVM
This projects aim at automatic regulatory road-sign detection with the help of distance to borders (DtBs) and distance from centers (DfCs) feature vectors. Developed system is able to detect and recognize regulatory road signs and the recognition system is completely based on the generalization properties of Support Vector Machine (SVM). The system consists of following processes: segmentation according to the color of the pixel, traffic-sign detection by shape classification using linear SVM and content recognition based on Gaussian-kernel SVM. Evaluation shows high success rate and a very low amount of false positives in the final recognition stage.
Face Recognition Technique Using PCA, LDA and Support Vector Machine
Feature representation and classification are two key steps required for face recognition. A novel method for face recognition is developed based on the combination of PCA (principal component analysis), LDA (linear discriminate analysis) and SVM (support vector machine). PCA and LDA combination is used for feature extraction and SVM is used for classification. The normalization had been done to eliminate redundant information interference previous to feature extraction. The experiments are tested using ORL face database and the evaluation section proves the robustness of the developed technique.