Far-infrared Pedestrain Tracking Method For Driver Assistance Detection Systems

Visual Tracking Technology required in many fields, such as Video Surveillance, Intelligent Car, Intelligent Transportation, Behavior Analysis and so on. Therefore Visual Tracking has become a hot research area. Secondly, in driver assistance systems, a nighttime pedestrian detection is necessary to reduce traffic accidents. It is a pity that all of existing detection methods in driver assistance have many problems, such as low detection rate, unstable detection. So it is very necessary to research nighttime pedestrian tracking method. By reading a lot of papers and taking simulation experiment,we proposes a pedestrian tracking method based on particle filter framework. The main research work and contributions are as follows: 1. The template update method based on image segmentation technology. In this paper, we use adaptive double threshold segmentation technique to obtain target which may contain pedestrians ROIs, then find target ROI by template matching method, and update the target template, So that the target template can timely follow up changes of target’s appearance and scale. 2. Diversity filtering particles. In the re-sampling phase, particle filter algorithm reserves particles having big weight, and discarding particles having small weight, then particles move towards high weight. But it also reduce the diversity of the particles at the same time. To restore the diversity of the particles, this paper make all of particles do Brownian motion to increase the diversity after re-sampling phase. 3. Improve tracking precision. This paper embeds the SVM classification into particle filter framework to filter most interfering particles, this process improves the predicting precision effectively. 4. Accelerate global search. This paper locate areas that contain target possibly quickly by extracting image texture information, then uses the PSO algorithm to search the target. This method avoids the global search process and improves the search efficiency greatly. The experimental results show that the proposed tracking method has better robustness and more accurate prediction than other tracking methods, and resolves the problem of short time target occlusion. Finally, it also has good tracking performance in tracking rapid moving pedestrians.