Traffic Congestion And Moving Target Detection Based On Machine Vision

With the rapid development of China's economy, the scale of urban vehicles growing rapidly, the resulting traffic problems in the city are becoming more and more serious and become one of the most important problems that plague the city's sustainable development and Intelligent Transport System provides a new way of thinking and methods to solve the urban transport problems. Using machine vision technology, which has the advantages of fast processing speed and high information content, to ease traffic congestion and improve traffic efficiency of road network, has become one of the hot topics in the field of Intelligent Transport System and a high theoretical value and practical significance.Based on the city transpotation video image processing and machine vision technology, this paper deals with the static and dynamic situation of the city transpotation under the Intelligent Transport System of "Nansha continuous flow transpotation controlling platform" and video traffic information intelligent detection system software.The main contents are:(1) Common image processing methods of traffic scenes;(2) Aiming at the defect of conventional sensors of high cost, difficult maintenance and low detection rate, an new algorithm for traffic congestion detection was proposed based on image texture features, which connects the traffic jam with image texture feature considers the dark light detection, automatically demarcates vehicle-zone according to the image color information, reduces texture feature extraction time;(3) Aiming at the defect in the Kalman filter Prediction and Mean Shift vehicles tracking algorithm of low tracking rate when the target is blocked or varies the speed, an improved algorithm combining both algorithms is proposed, which can distinguish the reason why the Bhattacharyya coefficient is small when target is blocked or suddenly changes in motions, thus it can improve the tracking efficiency;(4) Aiming at the defect in the HOG features + SVM's traditional pedestrian detection algorithm of unsatisfactory performance in complex traffic scenes, an improved pedestrian detection algorithm based on HOG- PCA + Adaboost structure is proposed, which reduces the dimension of feature vector using PCA, and can significantly reduce the machine learning time by using adaptive enhancement algorithms. Finally, the algorithm was developed under VS2010 and OpenCV circumstance, and the practical application has been found of significant effect and high engineering value.