Depth Image Based Object Localization Using Binocular Camera and Dual-stream Convolutional Neural Network

With the development of depth sensing devices, depth image based localization has received much attentions. In this paper, a depth image based object localization algorithm by the dual-stream convolutional neural network (CNN) is proposed. In the off-line phase, at each reference position, the grayscale image and its corresponding depth image pairs are collected by the binocular camera. The using the image preprocessing technique, the grayscale image and depth image are transformed to the three-channel images. Then, the dual-stream CNN with the shared weight coefficients are used for offline regression learning. At last, the distance based regression model is obtained. In the on-line phase, after the preprocessing of the grayscale image and depth image, the final distance can be estimated by the distance based regression model. Experiment are carried out to evaluate the performance of the proposed algorithm. The test results illustrated that the proposed algorithm has better localization performance than traditional image approaches.