Research On Multiple Targets Tracking Method Based On Multiple Information Fusion

With the development of intelligent equipment, people can gain vast amountsof information. And the most sensorial information is the visual information, whichmakes the computer vision become one of the important research contents in manyfields of research. In the field of computer vision, some researches can replace thehuman, and even more than the human. These subjects have been gradually paidattention to by many scholars. Multiple targets tracking task is a challengingresearch topic, and could have wide applications in the fields of video surveillance,robot vision, traffic management etc. Therefore, designing an effective multipletargets tracking algorithm is of great significance and value.But in the real complex scene, there will be the light changes, the targetquantity changes, targets occlusion between each other, targets cross and so on.These make the scene becoming complex and changeable in video sequenceacquired, so tracking multiple moving targets at one time will be very difficult. Withthe development of hardware equipment, more and more information sources canprovide a variety of information. This paper is committed to using the other sourcesof information besides the video image information, especially the depthinformation, in the multi target tracking task, and then finding an effective multipletargets tracking algorithm.The main contents are as follows:Firstly, in this paper we produce a depth information optimization algorithmbased on moving object detection. In order to introduce the depth information intomultiple targets tracking task, characteristics of the depth information are analyzedin this paper. And in order to make the depth information as useful as possible in themultiple targets tracking algorithm, combining with the detection results of multiplemoving targets position of the image, the depth image is optimized. The algorithmmade good preparation for using depth information in multiple targets trackingalgorithm.Secondly, we produce a multiple target tracking algorithm based on depthhierarchical segmentation. The algorithm first does some statistical analysis of thedepth information of the image which has been optimized. And then we find thediscrete distribution of the depth information, thus we segment the depth byhierarchies, and get regions in the depth, then track the multiple targets in eachdepth hierarchy regions respectively. So the number of targets reduces in eachhierarchy, and so the complexity of the scene does. Then combined the results ofmultiple targets tracking, the final result will be gained. Thirdly, we produce a multiple target tracking algorithm with multiple featuresbased on statistics of depth information in the regions. By analyzing the results ofmultiple targets tracking algorithm based on depth hierarchical segmentation, weuse depth information in another way. The depth information is not just segmentedinto hierarchies, and the regional values of the depth image will be used as thefeatures with other color features of the moving target. The effect of multiple targetstracking is enhanced.