Moving Object Detection And Tracking In Vehicle-mounted Video Monitoring System

The new generation of intelligent video surveillance technology has been widely applied to the field of public security monitoring, intelligent transportation, residential area surveillance and visual navigation fields. Vehicle-mounted video surveillance system is an important application direction of intelligent video surveillance technology. By vehicle-mounted camera, we can expand the scope of monitoring, continuously monitor and track the interest targets. However, the motion of car and the dithering of the camera caused by vehicle motion led to the dynamic movement of the monitoring background. In this complex dynamic scenes, moving object detection and tracking algorithms for static scenarios is invalid. So the study of efficient, robust and practical moving target detection and tracking algorithms has a significant research value for the complex and dynamic vehicle-mounted video surveillance scenarios.Firstly, this paper introduces the research status of moving target detection and tracking technology and related theoretical basis. Then, the paper introduces the overall architecture vehicle-mounted video surveillance systems, and details the algorithms used in vehicle-mounted video surveillance system and how to implement respectively. And the paper analyzes the technical difficulties of studying of algorithms in practical application scenarios. Since the vehicle-mounted video surveillance system in the background is in motion, a global motion estimation method based on Harris corner feature is adopted in the paper for background motion compensation calculation. After background compensation, the paper adopts a small footprint, low computation and high-efficiency operation of Vibe detection algorithm in target detection. Considering the problem of ghosting existed in Vibe moving target detection algorithm, this paper uses median filtering algorithm to improve Vibe algorithm. The improved Vibe algorithm can completely eliminate ghosting in a short time. After moving target detection, this paper uses morphological processing to optimize foreground moving objects making the detected moving objects more complete and accurate.In this paper, a new real-time compression Tracking (CT) algorithm in the moving target tracking detection is introduced, which is simple, efficient and robust with using very sparse random measurement matrix to obtain a low-dimensional projection of the class Haar-like features and using a simple Naive Bayes classifier to classify low-dimensional features. We combine perceptual hashing to improve the performance of CT algorithm from two aspects. On the one hand, we use a simple and efficient perceptual hashing modules to match a large number of samples screened searches, reducing the amount of computation of algorithm. On the other hand we use an adaptive parameter model update strategy to improve the adaptation performance of the algorithm under partial occlusion. The experiments shows that the processing speed and robustness of the real-time compression tracking algorithm with perceptual hashing algorithm has been a certain increase for vehicle-mounted video surveillance systems.