Obstacle Avoidance Based On Binocular Vision Smart Car

Smart cars mean that they may independently run without drivers under wild natural conditions and on various roads. Research on obstacle avoidance is a task of top priority and receives increasingly greater concern. It is possible to improve capabilities to smart cars of identifying road information by extracting the information to identify roadblocks with machine vision, especially binocular vision, in order to promote the application and development of smart-car technologies.In this paper, theories on smart cars were deeply examined and explored. Road information was converted into image signals with monocular image acquisition technology of machine vision, so as to obtain road information through image processing. Meanwhile, 3D road information was acquired with binocular vision to calculate the relative distance between a smart car and a roadblock, so that plans could be made for obstacle avoidance and ideas could be offered for actually realizing the goal of driverless cars.Based on the application of binocular vision in obstacle avoidance by smart cars, this study was primarily performed as follows:Firstly, actual road images were acquired with the binocular camera installed in a smart car and then processed. In consideration of unclear targets and backgrounds as well as defective backgrounds for targets in case of traditional Otsu method for image processing, an improved method for multi-threshold segmentation was put forward. Among multiple categories segmented by multiple thresholds, there must be a category of targets, while other categories should be integrated as background. Target and background w e differentiated with the ratio of maximum variance between clusters to within-cluster variance based on their difference. An experiment was performed on actual images and proves that the method was feasible.Next, the distance from a roadblock was er measured. A model of binocular vision was constructed in view of complicated calibration of a binocular camera and complex calculation. Based on the model, features were matched by SIFT to calibrate parameters required by the camera. This method was simpler than traditional calibration methods. To measure distance more accurately, corners of images were detected to extract sub-pixel coordinates of feature points. The simulation experiment suggested that the method mentioned in this paper was rational.At last, roadblock avoidance was analyzed.First of all, the size of the roadblock divided into two halves along the central line was calculated for a smart car. The roadblock was analyzed statically and dynamically. Subsequently, strategies for avoiding roadblocks were planned for statistic and dynamic roadblocks on the basis of measurements. Finally, experts’ inferences were introduced to examine roadblock avoidance of a smart car.