Depth Calculation and Object Detection Using Stereo Vision with Subpixel Disparity and HOG Feature

Object detection using image processing mainly includes single-camera based detection and stereo vision based detection. Stereo vision-based object detection approaches need to find correspondences from the left to the right image. In this paper, we proposed to detect the region of interest (ROI) of objects like cars, people, bikes using HOG descriptors and make correspondence matching only in ROI. Moreover, we used 3D reprojection method to calculate the depth and employed the subpixel disparity technology to increase the depth range resolution. By the proposed approach, the corresponding is only needed to be done for special features in the special region which improves the accuracy.