| The multi-target detection and tracking are becoming one of the hot issues in computer vision and have been popularly used and manifested with huge economic values in the fields of military, scene monitoring, human-computer interaction, etc. To solve the existing problems of target detection and tracking on the 2D color images, we fuse color information with depth information to effectively detect and track the targets under different circumstances such as illumination changes, targets appear, disappear and occlusion occurs.The main work of this thesis is illustrated as follows:(1) We analyze and summarize the current target detection and tracking methods based on color image or depth image. We design the 2D/3D camera imaging system and realize the registration between depth image and color image. In this way, we can obtain the depth image and color image from the same perspective.(2) We present a novel moving targets detection method by fusing color classifier and depth classifier. We establish the color and depth classifiers for each pixel and assign different weights for each output of the classifier adaptively to achieve effective targets detection. Experimental results show that the proposed approach works well in complex real-world scenes such as environments with illumination and shadow variances, the similar color or depth information of moving objects as the background.(3) According to the depth and spatial features of the detected moving targets, we compute the depth histogram of these foreground moving targets to adaptively determine the depth threshold interval. In this way, the targets are accurately located even when they are occluded in different depth layers.(4) According to the main problems of multi-target tracking such as objects occlusion, merging and splitting, we present a target association method based on multi-feature fusion of the detected and located moving targets. Then the multi-target tracking is achieved by analyzing the target association matrix. |
PDF Full Text Request