Research On Several Key Technologies Of Pedestrian Detection And Tracking

In the computer vision field, Pedestrian detection and tracking is a research hotspot and abasic task. It plays an important role in research and shows broad market prospects in trafficmanagement, intelligent monitoring, Human-Computer Interaction ect. The research of methodson pedestrian detection and tracking is of great importance. Although pedestrian detection andtracking technology has been studied for nearly20years, the effectiveness is not ideal as thehuman body is non-rigid and the most applied scenes are complicated. So far, pedestrian detectionand tracking is still an active area of research.On the base of having read lots of literature, the main research works are as follows:In the first chapter, introduces the application background, then elaborates research status,problems in pedestrian detection, and the content of this thesis.In the second chapter, introduces the relevant theoretical knowledge of pedestrian detectionand tracking algorithm. This chapter first states the common methods of describing thecharacteristics of objects, then expounds the basic work of imagery enhancement, and theAdaBoost algorithm and algorithm for SVM are described at last.In the third chapter, for conventional LBP operators considers only the size of the relationshipbetween center pixel and neighborhood pixels and regardless of their contrast, this chapterproposes an improved LBP description operator, which adds the contrast information. Theimproved characteristics of LBP operator are less sensitive to the image noise and light change.Combining with the SVM classifier, the experimental result shows that improved algorithm ismore ideal.In the fourth chapter,for the question of classifier degradation and imbalance betweenpositive and negative samples. This chapter first updates the rule of weights in the AdaBoostalgorithm, then normalizes the weight of positive and negative samples. Finally, on the basis ofAdaBoost algorithm, this chapter brings in the SVM classifier which is in the cascade form,andcan remove the false detection of AdaBoost.In the fifth chapter, proposes an efficient robust pedestrian tracking algorithm. The tracking algorithm based on the original mean shift algorithm displays the phenomenon of tracking errorwhen background color is similar to the target color information. To solve this problem, thischapter introduces the color characteristics of spatial information into the Camshift algorithm, andthen determine whether update the target template or not according to the template update factor.Experimental result shows that this algorithm can track target accurately, stably, and in real-timeunder the background disturbance.In the sixth chapter, summarizes our works in the thesis. Shortcomings of works were pointedout and the objectification research of facial diagnosis was prospected.