| With the rapid development of highway transportation in China, road traffic accident has become one of the most serious problems that threaten the safety of public property. According to analysis of road traffic accident statistics annual report, large vehicles especially passenger vehicle accident is the main cause of major traffic accident, it has a very baneful social ramification while the dominant factor of major traffic accident vehicle is the drivers’ dangerous driving behavior. With the development of science and technology, more and more safety and assistant driving equipment has been popularized to all kinds of small vehicles, but the vast majority of passenger vehicles have not yet been fitted with safety assistant equipment. Therefore, it is significant to carry out the research on the lateral and longitudinal dangerous driving state identification technology to provide technical support for the active control and participant management of passenger vehicles, which has great influence on improving passenger vehicles’ safety and avoiding casualties and economic losses.Relying on the National Natural Science Foundation of China(51278062), this paper applied a combination of computer graphics, traffic engineering, vehicle engineering, artificial intelligence & other multi-disciplinary theory and CCD vision sensor acquisition technology, digital image processing technology, machine learning and pattern-recognition technology, multi-source information fusion technology, swarm intelligence technology & automatic control technology. Through a lot of theoretical modeling, static off-line experiments, dynamic real vehicle experiments and massive data analysis, the horizontal dimension and vertical dimension of passenger vehicles in real-time online identification of dangerous driving state was analyzed. The program can timely warn motorists about abnormally dangerous driving behavior with state recognition technology and the related system was finished.In view of the difficulty of balancing the contradiction between robustness and real-time performance of road environment information detection algorithm and the problem of high false alarm rate in traditional vehicle lateral departure identification model, a multi-thread visual feature constraint was adopted to solve the road equation and vehicle lateral departure rate identification. By enhancing the gray scale of road pictures, modifying the median filtering denoising, extracting directional filter lane edge, searching the dynamic region of interest and Improved optimal threshold segmentation for sequence image to excavate contour information of the lane;Based on the multi feature set lane marking line selection, combined with the width of lane marking line, the least square method was used to realize the detection and location of road marking line;The Kalman filter tracking model was introduced to improve the detection efficiency of lane marking line and the ability of anti-jamming, which can balance the contradiction between the robust and real-time performance of lane detection. The key information of the road was reconstructed based on the inverse perspective projection transformation, and the running state of the vehicle was detected in the world coordinate system, Taking advantage of the advantages of low alarm rate based on spatial information early warning mode and the timely warning based on time information, a vehicle lateral departure identification model based on temporal and spatial information fusion was established, and the effectiveness of the warning system was improved.Aiming at more interference factors in the process of the front vehicle identification and single consideration for longitudinal distance, the method of vehicle infrastructure cooperative was taken to research the identification of longitudinal dangerous driving state of the front vehicle. Based on massive amounts of offline training sample set, effective vehicle contour and texture characteristics was extracted, Haar–like characteristics was used to describe the goal, Adaboost machine learning algorithms was used to trained classifier, the sample characteristics of cascade classifier was built, and the test object was used to detect the vehicle existence; Making full use of the multi-hypothesis ability of particle filter, the rapid and stable tracking for front target vehicle was achieved from the local optimum perspective, it is stronger adaptive for the uncertain factors such as environmental interference and vehicle type; Based on the key parameter accurately calibration of the CCD visual sensor, and condition of dynamic compensation algorithm for error of measurement on the inherent characteristic parameter variety of the front vehicle suspension, the longitudinal distance measurement model driven by lane plane geometry model was built to realize the precise measurement of the vertical distance. Fully considering the driver cognitive response characteristics, vehicle response characteristics and road environmental and so on, swarm intelligence technology was adopted to construct the vehicle longitudinal driving safety domain and the numerical simulation was analyzed; Fusing the front vehicle location, the road-vehicle identification model of longitudinal dangerous driving state was established and early decision-making warning was made through the time domain risk model, to ensure both the driving safety and road traffic capacity at the same time.Aiming at high using-cost of the current real-time acquisition system of vehicle driving state parameters and can’t be compatible with the vehicle lateral &longitudinal dangerous state key information detection system in the paper. The real-time data acquisition during the driving process was studied based on hybrid heterogeneous computing platform. Under the microprocessor platform, the real-time acquisition minimum system of vehicle driving state parameters was built; Active oscillator was used to decrease the interference generated by the oscillator on other circuit; Optocoupler isolation circuit was used to suppress the interference of the speed signal pulse source, and the vehicle driving state parameters acquisition system which embodies the task allocation of self organization was introduced to improve the robustness and the coordination of the parameters. The parallel design strategy to system-tasks-module was proposed, and ECT task module accumulator was utilized to account rising edge of vehicle speed signal. RS232 serial communication protocol was used as seamless connection between the real time acquisition system of vehicle driving state parameters and the key information detection system of the lateral & longitudinal dangerous driving state, so as to solve system platform running compatibility issues, and improve the vehicle running state parameters collection’s accuracy and practicality.In order to verify the correctness and effectiveness of the proposed algorithm, the system platform based on the upper computer module was built and the key hardware selecting work was done. Based on the experimental platform of the lateral & longitudinal dangerous driving state key information detection system, the experimental images and data was used to verify the related algorithm proposed in the paper. The results showed that the lateral & longitudinal dangerous driving state key information detection algorithm proposed in the paper was effective and feasible; the system worked stable and reliable, and met the design requirements. |
PDF Full Text Request