Pedestrian Detection Technology With Partial Occlusion

With the continuous updating and development of computer technology, Pedestrian detection technology has shown important application value in intelligent video surveillance, human-computer interaction, intelligent transportation and other fields, has been a hot research topic in computer vision field. But because of the occlusion problems, pedestrian detection with partial occlusion has been a difficult problem in this field. State-of-the-art methods perform well in secens with relative few people, but it is the fact that current people detectors fail when persons are strongly occluded. In this paper, the pesestrian occlusion as the main research direction, based on deformable part model, we propoes a joint detection model with different degrees of occlusion problem. the main research contents are as follows:(1) The occlusion problem is subdivided into two categories of object occlusion and pedestrian overlap occlusion by the detailed outline of the occlusion problem. In order to better evaluate the detection effect and the training of the later models, the UESTC occlusion dataset is established.(2) By analyzing the HOG features,features pyramid, model training and detection process of DPM target detection algorithm, we trained a single person occlusion model to detect pedestrians.(3) According to drawbacks of single person model in pedestrian overlap occlusion problem, we propose a two person model detector and a joint person detector that includes both a single and double person model and test the problem of different degrees of occlusion.(4) By experiments on INRIA and TUD-pedestrians two public available datasets, it is confirmed that the joint person model can improve the detection rate of pedestrian compared with the single person detection model.The experimental results show that the joint person model can better solve pedestrian detection with occlusion handling.