Research On Traffic Flow Measuring Using Video Detection Algorithms

Intelligent Transportation System (ITS) has been considered, by many scientists, as the most effective way to solve the urban traffic problems. In the meanwhile, it is the main stream direction of urban traffic development. Acquiring the real-time information of traffic, such as traffic flow, vehicle speed, is the foundation of ITS. In this paper, we do research on the approach of traffic flow measurement using video detection algorithms in ITS, and pay more attention to the key algorithms that related to this approach, such as moving vehicle detection, shadow elimination and vehicle counting. Based on the result of research, a simulation system is developed to realize traffic flow measurement. This paper consists of four main sections.1. Moving vehicle detection and segmentation. Summarize and analyze the current algorithms of detection and segmentation, then compare them through specific experiment and enumerate their advantages and disadvantages. Aiming at overcoming the shortage of these algorithms, an improved method, combined inter-frame difference algorithm with ViBe, is proposed. Concrete contents:reviews of foreground object detection algorithms. In this paper, Mean method, Median method, Mixture Gaussian Model, W4 Method, Eigen method, Nonparametric Kernel Density method, Codebook method and ViBe method are exploited to detection the foreground object. The results of experiment show that Vibe method is of higher correct detection rate and faster processing. However, the problem of high false-positive rate also exists in this method; Algorithm Improving section. In this section, an improved method is proposed to reduce the false-positive rate. This method combines inter-frame difference algorithm with ViBe method so that the ghost in the first several frames can be taken into background model. The experimental results demonstrate that the proposed method not only keep the advantages of Vibe method but also decrease the false-positive rate. 2. Shadow detection and elimination of moving vehicles. Describe the importance of shadow removing, then detect and eliminate shadow from spectral point of view. Specific content:this paper describes the principal of generating shadow, as well as, the shadow features. List the classification of shadow detection. Review the principle of shadow detection based on color space and on texture respectively. Compare the four algorithms of shadow detection through concrete experiment. Draw the conclusion that shadow detection algorithm based on HSV color space has a higher correct detection rate and low false-positive rate.3. Vehicle counting. A simple method is used to count vehicle, based on the result of previous processing. Specific content:adopt the mathematics morphological operation to process the binary image aiming at improve accuracy of the following algorithms; Compare the two vehicle counting methods—region labeling algorithm and virtual loop detection algorithm. Draw the conclusion, based on the results of experiment, that vehicle counting method based on virtual loop represents the performance of simple, real-time.4. Overall system implementation. Develop a Traffic Flow Measurement software system to verify the algorithms, proposed in this paper, performance of accuracy and real-time. Two real traffic scene videos are processed, the result illustrate that the system in this paper has high detection accuracy, in addition, the average speed of processing the two videos can achieve about 27fps and 18fps respectively, which fully meet the real-time performance of traffic information collection requirements.