Research On The Intelligent Vehicle Reversing Control Based On Binocular Vision And Sparse Representation

In recent years, with the increasing vehicle number and the high accident rate, the vehicle safety problem has become one of the most important issues in Chinese economy and society. Therefore, to develop the intelligent vehicle safety technology is a very important role in reducing vehicle accidents, improving social security, and promoting sustainable economy in China. In this paper, we present a vehicle safety method for reversing speed control based on obstacle detection and sparse representation. The studies included in this paper are summarized as follows:First, we present a vehicle safety framework for reversing speed control based on obstacle detection, sparse representation and analyze the binocular vision theory. With using binocular vision sensing technology, we obtain the 3D-depth information of the object in the rear of vehicle. Calibration as a part of the preparatory work for measurement is used to get the intrinsic matrix, distortion vector, rotation matrix, translation vector, and some other parameters so that we can get a pair of images that are nearly coplanar. Then, stereo rectification, namely, removing distortions and turning the stereo pair of images into standard aligned form by utilizing the calibration results, is an important work for calculating the disparity. Stereo matching is used to find corresponding points in the stereo images and calculate the disparity. Finally, after stereo calibration, rectification, and matching, triangulation is utilized for computing the position of objects in the real 3-D space.Second, do the preprocess work of the binocular image before tracking. After obtaining the disparity and 3D-depth information, denoising and segmentation to the object image must be done. We use the dilation algorithm to connect discontinuous region and then use erosion algorithm to remove small noise points. Use a threshold value to binarize the preprocessing disparity image. The threshold controls a security distance that can classify the disparity into two parts. One is the close obstructions area and the other is all background area. The region labeling algorithm is used to mark different obstacles in different area.Third, analyze the tracking and recognition algorithm of sparse representation. Tracking is another part that is as important as the previous 3-D depth acquisition in the vehicle reversing system. However, a tracker has difficulty in performing very well in both robustness and real time. Here, we present a sparse representation to accomplish the tracking and recognition task. The mainline of the algorithm is:using the particle filter based on the Monte Carlo method as the structural framework. Meanwhile, Li minimization is used to solve the sparse representation problem for tracking and recognition. Our obstacle tracking has achieved a good performance by particle filter and sparse representation.At last, establish vehicle dynamics model and analyze the vehicle reversing control algorithm with simulation. When a driver starts reversing, ECU gets a reversing signal and the binocular-cameras automatically open to receive images. Obstacles behind the vehicle are detected and tracked using binocular-cameras algorithm and sparse representation. Then ECU makes an operation decision based on the obstacle distance and acceleration pedal condition. We use the Matlab Simulink to establish vehicle dynamics model, and then, use the fuzzy control to achieve speed-distance curve simulation, by which the final performance evaluation demonstrates the validity of the proposed vehicle reversing speed control. In the end of this paper, we also proposed a multi-information fusion reversing control method to discuss the related future work.