Semantic Analysis Of Vehicle Behavior In Surveillance Video

As the rapid development of the current economy, and the increase of population and road vehicles, leading to the traffic problem is increasingly serious.To effectively solve this problem the Intelligent Transportation system(Intelligent Transportation Systems, hereinafter referred to as ITS) arises at the historic moment. Through processing the computer video data, acquiring automatic scene information, semantic analysis interest goal to realize intelligent traffic monitoring system is a topic worth studying.In this paper, first of all,do in-depth research on the premise work of moving target detection and tracking, and propose fast algorithm of the gaussian mixture model for the target detection algorithm,and then propose improved tracking algorithm that based on kalman predictor for target tracking, monitor video semantic analysis part.First proposed the vehicle behavior analysis algorithm that based on trajectory tracking, and adopt the machine learning algorithm SVM to analyze vehicle behavior in video.This paper mainly completed the following work:Study the current commonly used moving object detection algorithm, propose rapid mixed gaussian motion target detection algorithm which based on gaussian mixture model.First quickly determine the target area by three frame difference, and then merely to match the gaussian mixture model of undetermined target area, determine the target area and the critical area, adaptive areas set background update rate, finally use the feature space to shadow suppression(R, G, and I).The experimental results show that this algorithm can fast detect moving targets, and can overcome the influence of shadow to a certain extent.Study the commonly used moving object tracking algorithm.In the base of the traditional camshift tracking algorithm, improve it be based on kalman predictor.First of all, according to the target detection results search window initialization, Bhattacharyya distance judgment whether keep out by calculation, in case of occlusion kalman predictor as a result, the end of the block to camshift tracking, the experimental results show that the algorithm can adapt to a certain condition, effectively improve the traditional camshift tracking results.Use the adaptive piecewise linear least squares fitting method to fast tracking trajectory fitting, to extract the motion parameters according to the results of the fitting speed rate and direction change rate vehicle behavior model is set up, implement unusual behavior of vehicle detection based on tracking trajectory, the experimental results show that the algorithm can quickly and efficiently detect the traffic surveillance video vehicles abnormal behavior,such as turn, the brakes and the brake turn.Through researching machine learning algorithm,chooseing adopt SVM(support vector machine)to analysis of vehicle behavior in surveillance video, first determine road vehicle four behavior patterns,as turn left, turn right, stop,go straight of training sample set, using RBF kernel function training samples for SVM classifier.The experiment shows that the algorithm can carry on the conduct semantic analysis to the intersection traffic surveillance video, analysis of accuracy to 90%, and machine learning method has a higher initiative.