Research On Moving Target Detection Of Traffic Control System Based On CUDA Heterogeneous Platform

Transportation is an essential part of people's life,and people become to pay more attention on intelligent monitoring of road condition and safety transportation.Using intelligent monitoring allows monitoring system detect moving object automatically,automatic tracking and behavior understanding.As a first step of video frame image processing in intelligent monitoring system,motion target detection plays an important role in the late tracking identification and has research and application value.In the static background,background subtraction and frame difference method are commonly used moving objects detection methods.Under the background of dynamic,the commonly used moving object detection methods such as optical flow method.The traditional moving target detection algorithm is based on the calculation of CPU,it has long running time and low efficiency shortcomings and can not meet the real-time requirements generally.In order to measure the radial movement of the target vehicle accurately,reducing the noise of the system and elevate real-time of systems,this paper proposed a new algorithm for moving object detection based on improved parallel frame difference and sparse pulse coupled neural network(PCNN)by using parallel computing architecture(CUDA).Because the research object is moving vehicle,the road condition is more complex and have more collected original noisy image,this paper proposed a kind of filtering algorithm which combine with average filtering and histogram transformation on image preprocessing.Because of the low target detection rate of traditional algorithm on rapid movement,in the process of this topic research we according to classic algorithms principle such as interframe difference method,optical flow method,background difference method,combined with the characteristics of wavelet transform,this paper first put forward the second three frame differential method combined with multiple small scale flow method of moving object detection algorithm.This kind of algorithm can detect fast moving target accurately in complex dynamic background,but in the detection of the radial movement the detection rate was a little lower.To solve these problems,combined with PCNN mathematical model and work principle this paper proposed a new algorithm of moving object detection based on improved frame difference and sparse PCNN.According to the characteristics of the video frame image processing,combined with the independence of the pixel units,considering the parallel structure and hardware features of graphics processing unit,The improved frame difference method to get the two value image process,and the difference between the two value image mapping to the sparse PCNN model are put on the GPU,and choose grain storage and shared storage way to improve the efficiency of data access and reduce the complexity of the algorithm which achieved the combining of GPU and CPU algorithm and improved the execution efficiency of algorithm.In order to verify the validity and real-time of the algorithm,this paper going through VS2010?OPenCV2.4.9?CUDA to contrast the algorithm and related experimental and finally verified the accuracy and real-time performance of moving target detection.