Research On Object Detecting And Tracking Under Complex Environment

Object detection and tracking is an important research field of computer vision, which is widely used in civil and military fields. Object detection is to identify the target area from the image by the analysis of image sequences. Object tracking is to predict the position of target in the next frame by analyzing the target's data of the image sequences, thereby generating the target trajectory. In recent years, with the development of computer technology, object detection and tracking has become a hot research field. Scholars have presented a lot of algorithms in this field, but none of them can be applied to any scene due to the complex contextual issues, such as illumination variation, dynamic background, distortion of moving object, scale transform, occlusion and so on. The research of object detection and tracking algorithm is still of great significance.In terms of object detection, this thesis focuses on two kinds of detection methods,including detection methods of motion analysis and detection methods based on statistical learning. Detection methods based on motion analysis mainly include frame difference,background difference, optical flow method. Experiments can confirm that this kind of methods can detect moving targets, but it is limited to relative movement between target and background scene. Detection methods based on statistical learning mainly use the classifier technology. Random ferns classifier using 2bitBP description is adopted in this thesis, and random ferns classifier is used to object detection in the aerial video through the sample collection, classifier training, classifier test and other steps. Experimental results show that the random ferns algorithm has high detection rate and high adaptability in complex backgrounds.In terms of object tracking, this thesis improves the particle filter tracking algorithm.HOG description and color description are combined as the target description, which can better represent the target in the tracking process. In order to improve the robustness of tracking and cope with occlusion, the target model based on space-time context information is proposed in this thesis and the update program is improved. For the object scale transform during the tracking process, a new method of scale change is presented. In terms of object lost condition during the tracking process, we proposes an improved heavy detection method that is based on the TLD tracking algorithm.Finally, we conduct some experiments to verify the tracking algorithm. By comparing with the traditional particle filter tracking algorithm, we assess the tracking algorithm in this paper qualitatively and quantitatively. Firstly, the tracking performance of two methods in three representative image sequences is shown in the thesis, then, center error map and overlap rate map are drawn by calculating the center error and overlap rate of two tracking algorithm. Experimental results show that the improved tracking algorithm in this paper performs better in object tracking in complex background.