Research On Pedestrian Detection Algorithm Based On Codebook Background Modeling

Research on pedestrian detection is essential for the protection the safety of pedestrians, most of the current research is the detection of pedestrians on a specific data set, but for the pedestrian detection under the particular scene, like the substation, it should be designed on characteristics of the scene, gives full performance of the algorithm.This paper analyzes the characteristics of the substation scene, according to features of the fixed scene and less pedestrian in the scene, after selecting Codebook background modeling algorithm to roughly detect and obtain prospects, then using pedestrian detection algorithm to fine detect pedestrians in the foreground, the pedestrian detection task to be finalized under such scenarios. Throughout the algorithm process, mainly doing the work in the following areas:Firstly, research on background updating steps of Codebook background modeling algorithm, propose the use of temporary block model to update the background model, and solve the sensitive issue of Codebook background modeling algorithm to light. While the results of pedestrian detection also be incorporated into the background update process, let temporary block model will not update pedestrians which have the background characteristics into the background model.Secondly, improve pedestrian detection performance in both the classifier and search strategies. Compared to using a linear kernel SVM training pedestrian classifier, histogram intersection kernel makes the SVM has a more powerful classification capabilities without bringing significant computing consumption; analysis found that it is the bottleneck of traditional pedestrian detection method to search for pedestrians through sliding window to traverse the whole picture, so taking the coarse-to-fine manner, first using Codebook background modeling algorithm to quickly extract the foreground region, then using the trained pedestrian classifier to detect pedestrians in the foreground region through sliding window, in order to achieve real-time pedestrian detection.Finally, simulating the scene under the substation and the environment interfere, and testing the designed algorithm. Experimental results show that the improved algorithm can cope with the effects of environmental change, improve pedestrian detection performance, the average detection time of the Algorithm in the test video is about 29 ms, to achieve real-time detection requirements, and implement personnel intrusion detection under the substation scenarios.