Research Of Vehicle Detection Technology Based On Video

Intelligent transport system (ITS) is based on the digital image processing technology, its aim to analyze the video image sequences collected by the camera, detect and track the moving targets, then analysis and evaluation of the behavior of the object to gain more advanced understanding of image information, moving target detection and tracking is the key technology of intelligent transport system, and determine overall performance and practical value of the system. With the continuous improvement of real-time and accuracy, Intelligent transport system will be widely used in the future.The main research topic is based on the video monitoring system of vehicle detection technology. The main contents include two parts which are the foreground target detection and moving target tracking and statistical counting.At first, the paper compare and analyze various commonly used motion detection method, then proposed a background model based on adaptive filtering method, in which use the multiple Gaussian distribution model to simulate each pixel changes of the images. Gaussian mixture model approach to build up models for both foreground and background simultaneously and adaptively updates the background image. This algorithm have better results in separating foreground targets and suppresses background noise, good adaptability to the light changes and slowly moving targets.In addition, this article analyzed the basic principle of the Kalman filter, and proposes a tracking algorithm of multiple moving targets based on Kalman filter. Using the advantages of Kalman filter, this algorithm can estimate the location of moving target of next frame image in the basis of previous measurement data, and thus reduce calculation time effectively of the image searching and matching, and improve the accuracy of vehicle tracking.And based on the vehicle tracking, this paper proposed a vehicle statistical algorithm based on the virtual detection region. In order to achieve video-based real-time statistics of vehicles, we use the improved background subtraction method to region of interest (ROI) in the images, and combine with image morphology operations and shadow suppression method. This method has better adaptability to illumination changes in the environment, and has no restrictions in lane numbers and vehicles movement state, and has better performance in real-time scene adaptability with high accuracy.