This thesis investigates the effect of feature descriptor selection on pedestrian detection process. Pedestrian detection, the problem of... Show moreThis thesis investigates the effect of feature descriptor selection on pedestrian detection process. Pedestrian detection, the problem of recognizing, locating and tracking humans in videos has become a challenging task in computer vision area. Video surveillance and advanced driver assistance systems (ADAS) are few appli- cations of pedestrian detection. At the very rst stage of the pedestrian detection system, feature descriptors are created to describe the texture, shape and color in- formation of the pedestrian candidate. These extracted features, acts as an input to develop and test a pedestrian classi er Feature descriptor techniques su er from di erent kinds of problems such as illumination variation of the dataset, pedestrian pose di erences and noise in the region of interest (ROI). To overcome the problem of illumination change, depth maps, can be exploited due to their robustness under various illuminations. Considering real-time nature of the targeted applications, another important challenge for the feature descriptor technique is the computational complexity and also the resulting dimension of the feature space. In the initial part of this thesis, several feature descriptor techniques are ex-plored. In the primary part of this thesis work, two critical stages of pedestrian detec- tion are improved, depth map computation stage and the feature descriptor creation stage. Two novel fast techniques to compute depth maps are proposed for the depth- map computation stage; three newly proposed new descriptors are proposed which work on depth channel or a combination of depth and color to adaptively overcome the illumination variation problem while still maintaining low dimensional feature space. These methods are found to outperform existing state-of-the-art descriptors and with exceptionally low computational complexity. M.S. in Electrical Engineering, May 2015 Show less