Cheng Senlin

Fuzzy evaluation of traffic flow stability based on the discreteness of traffic parameters

The microscopic traffic flow stability is difficult to describe the influences of the dynamic relationship between vehicles on traffic conditions. In order to solve the problem, from the perspective of macroscopic and microscopic parameters of traffic flow and considering the discreteness of velocity and following distance, the paper selected average velocity, average following distance, the variation coefficient of velocity and following distance as the traffic flow stability indices. And a stability evaluation model was constructed based on fuzzy theory. Using VISSIM simulation software, simulation experiment was carried out. The influences of traffic volume, traffic incident, traffic signals and GPS permeability on traffic flow stability evaluation were analyzed. Results show that traffic flow stability decreases gradually with the increase of volume. When there is a traffic incident, traffic flow stability decreases at the upstream section, while it is contrary to the downstream section. With the decrease of GPS permeability, the differences of the four indices and traffic flow stability at these GPS permeability with that of full permeability are becoming bigger and bigger. So the behavior described by the model is consistent with the actual state of the traffic system, proving that the model is real and effective. Thus it can be used to evaluate urban traffic flow stability macroscopically.

An extended FVD model considering road radians and gradients

The microscopic car following model can depict the interaction between adjacent vehicles, but the existing studies rarely consider the effect of complex road conditions on the model. Based on the FVD model, an extended car following model with the consideration of the road radians and gradients is proposed in this paper. The linear stability condition is obtained by applying the linear stability theory. Experiments in different scenarios are carried out by VISSIM to evaluate the model performance. Results show that the extended FVD model can accurately reflect the driving behavior under real road conditions.

Highway traffic abnormal state detection based on PCA-GA-SVM algorithm

In the highway traffic abnormal state detection, Support Vector Machine (SVM) algorithm is widely researched in recent years, but it still has some limitations. Aiming at the problem of improper selection of feature vector, the space and time characteristics of highway traffic abnormal state data is summarized, and the feature vectors of SVM are selected by Principal Component Analysis (PCA) properly. To solve the model parameters selection problem, the theory of Genetic algorithm (GA) is used to select SVM model parameters effectively. Also two-class SVM classification is extended to multi-class SVM classification which has a better command of the traffic running state on highway. According to the severity of the traffic incident, the traffic is divided into the state of no event, the state of mild congestion and severe congestion. The test software of SVM algorithm and the improved one are developed by using MATLAB and LIBSVM tool. The experimental result shows that the improved algorithm has a higher accuracy and a higher detection rate.

Corner point set-based camera monitoring range deviation detection method in video surveillance system of highway

Camera monitoring range deviation detection is the prerequisite and foundation for monitoring video content analysis. Due to the interference of moving object, light and noise in the scene of highway, the existing detection methods have some problems in aspect of real-time and robustness, which cant meet the demand of highway monitoring. This paper presents a new method to detect camera deviation based on corner set feature. The proposed method eliminates pseudo-corner points based on the extreme value of Taylor series at corner points and the low clustering property of the random noise. Then the corner set feature obtained by training is adopted to represent the image accurately. In addition, cross correlation and dynamic threshold analysis method are applied to correct the event falsely detected as a deviation. Finally, the contrast experimental results demonstrate that the proposed method can improve the real-time performance as well as ensure a high detection rate, which meets the requirements of highway monitoring in practical application.