Detection System Of Ahead Vehicle Based On Data Fusion Of Ranging Radar And Vision

Increasing cars make traffic accident occur frequently, which brings great loss of life and property. Vehicle collision avoidance system can reduce or avoid collisions and improve driving safety, which becomes automotive manufacturers and scholars research priorities. Vehicle collision avoidance system is an advanced vehicle safety driving techniques, which includes driving environment perception, vehicle dynamics modeling and control, traffic safety distance model and decision-making, and so on. In this paper, driving environment perception is studied, and vehicle detection system in front is researched and developed. Using a variety of sensors mounted on the vehicle to collect environmental information in front of vehicle. Sensor data processing algorithms are researched. Multi-sensor fusion algorithm is researched. Use these algorithms to select effective target vehicles that pose a threat to the current vehicle, and get accurate relative distance, relative velocity, relative acceleration and other status information. It is important for this system to improve vehicle collision avoidance system performance and reduce collisions occur.The main research work of this paper includes:1. In this paper, the overall design of vehicle detection system in front is introduced firstly. Analyze the functions of detection system. According to the system performance needs, select types and quantities of sensors reasonably. Describe research ideas of each sensor’s data processing algorithms and multi-sensor data fusion algorithm. Identifies the key technologies required for achieving system functions and their relationships.2. Target recognition and tracking based on millimeter-wave radar(MMWR) is studied. Analyze a variety of actual conditions of MMWR working and corresponding characteristics of data. According to the design specifications of the road, select the closed target in the same lane as primary target. Between two adjacent data acquisition cycle, target physical motion is limited in a certain limits. With this physical fact, the real-time multi-target tracking algorithm is designed. Use this tracking algorithm to perform consistency test to effective target. By setting the threshold of life for the effective target, decision-making method of effective target is established, which could adapt to a variety of conditions to maintain the original effective target or switch to a new effective targets. Eliminate false targets, data-missed phenomenon, bumps and yaw of the vehicle, and so on, which can interference effective target selected. Then obtain continuous measurement data of effective target. Use fourth-order classical Kalman filter to estimate the motion state of effectively target. Taking into account the statistical properties of system random noise that is unknown and variable, Sage-Husa adaptive Kalman filtering is used to improve accuracy. Propose a method to prevent the filter divergence. Finally, this algorithm could reduce estimating errors and improve accuracy and real-time tracking performance.3. Haar-like rectangle feature of the image is introduced. Use integral figure to calculate rectangular eigenvalue quickly, and obtain eigenvalues vector. Based on AdaBoost algorithm, extract several effective features from the rectangle eigenvalue vectors. For each feature, build the corresponding weak classifier. And according to classifying accuracy level of each weak classifier, combine these weak classifiers into a strong classifier using different weight. Finally, a cascade classifier is established.4. Establish multi-sensor information fusion model is studied. Analyze the pinhole projection principle of camera and its linear model. Establish a linear conversion relationship between the camera coordinate and pixel coordinates. Taking into account the distortion phenomenon of the actual lens imaging, perform the nonlinear distortion correction for ideal linear model, and obtain corrected nonlinear coordinate conversion relationship. Because of the relative position between MMWR and camera is fixed and known, radar coordinates conversion relationship between MMWR and camera could be established. Finally, build a multi-sensor data fusion model in space. Because of the different working frequency between radar and camera, regarding radar data as a benchmark, use Multi-thread synchronization programing method to realize multi-sensor data fusion in time.5. At the end of this paper, hardware and software of vehicle detection system platform is designed. Integrate the radar data processing methods, multi-sensor fusion model and the trained classifier into the system software platform. Perform experiments. The results show that the detection system is better for achieving system functions.