Research On Traffic Congestion Detection Based On Machine Vision In Lightweight

^the problem of traffic congestion in urban become more and more serious, it is of significance to find a fast and efficient solution for traffic congestion detection. The miniaturization of detection system is needed for the high cost, the difficulties in maintenance, large amount of computation and high complexity existing in the current traffic congestion detection systems. In order to solve the issues mentioned above, a road traffic congestion detection approach based on machine vision in lightweight is designed and implemented in this thesis.The main research work and results are as follows:1.Following the idea of "points stand for a plane ", road areas and their sampling points in these areas are customized, and then resources for computation and storage can be reduced significantly. Thus, the machine vision-based algorithm for traffic congestion detection can be applied into embedded devices, and the lightweight automatic vision-based detection for traffic congestion can be realized in the front-end.2.Given that the integrity and the simplicity are demonstrated in the process of human’s observing the road, the road is taken as the focus of the research, and the congestion detection method utilizing the sample points are proposed in the thesis. The algorithms of background subtraction and inter-frame difference are combined to obtain the non-existence sampling points, the existence sampling points, the motion existence sampling points and the motionless existence sampling points, respectively. Then, the key traffic parameters, such as lane occupancy rate, vehicle queue length and vehicle flow can be obtained by considering the spatial distribution of the existence sampling points. Moreover, the state of traffic congestion can be inferred by analyzing the spatial distribution of the existence sampling points and the motionless existence sampling points in lane area. The proposed method is proven to be more intuitive, simpler and less amount of computation. Furthermore, it is independent of the traffic parameters.3.The advantages and disadvantages of RGB and HSV color model, which are used in the road background modeling, are fully analyzed. Due to that the background pixels on the road can reflect the environmental changes, an adaptive background modeling approach based on sampling points is suggested in the thesis. The problem that foreground pixels may be updated as background ones in the traditional adaptive background modeling methods is solved here by selecting different update strategies for the background pixels and the foreground ones.4.To mitigate the interference from the shadow of vehicles, a kind of shadow elimination method based on HSV color model is recommended and realized with the help of the characteristics of HSV color model. And the efficiency of the method is demonstrated by ideal experiment results.5.The application system is designed and implemented. The system consists of the detection node and the control node. The detection node, which is realized in the embedded devices, is used to capture the video images and evaluate the congestion states. The control node, which is realized in a specific computer, is in charge of initializing the reference background image of the road and displaying the results.