Online multi-target tracking via depth range segmentation
Hongyang Yu, Lei Hua Qin, Qingming Huang, Hongxun Yao, Liang Li · 2017
Recently, visual tracking with RGB and Depth (RGB-D) data has attracted increasing interest. However, the depth data sometimes is incomplete, since the objects are beyond the distance limitation of sensors cannot be sampled. Traditional depth based features, such as objects shapes in depth image, are no longer effective in this case. To solve this problem, in this paper we propose a depth enhancement method and introduce the divide and conquer idea into multi-target tracking. Firstly, we enhance the depth image by moving object detection, and then segment depth range into several small regions according to the depth range levels, which is discovered by our statistical model over depth data, so that the number of targets and interfering factors in each region is greatly reduced and hence the tracking in each region becomes much easier. Experimental results on a set of challenging sequences validate the effectiveness of the proposed method.