Research On Micro Intelligent Vehicle And Its Key Technologies Of Vision Perception

With the development of sensors, visual perception and artificial intelligence techniques, driverless intelligent vehicle had appeared in front of people quietly. Currently, driverless intelligent vehicles were mostly equipped with expensive and high-end sensor system, far more than the fair value of the public can accept. At the same time, conducting experiments and tests on driverless intelligent vehicles under real-world scenarios existed constraints on security, policies regulations, cost for development and other factors. The micro intelligent vehicle simulation platform was able to simulate the actual driverless intelligent vehicle test on the extent and provided a convenient debugging environment for unmanned technology. Therefore, we relied on the existing carriageway co-simulation platform of "Shaanxi Provincial Road Traffic Intelligent Detection Equipment and Engineering Research Center" in Chang'an University and upgraded hardware system of the existing micro intelligent vehicle based on the specific needs of driverless missions. This article used inexpensive vision sensors to sense the traffic environment, focusing on the visual perception of the key technologies involved driverless. It provided not only a theoretical basis for driverless real tests, but also a useful attempt to reduce the cost of driverless vehicles.Firstly, we conducted a research on the micro intelligent vehicle system. It mainly related to main processing system, sub-control system, drive system, steering system and visual system. We upgraded the host processing system hardware configuration and improved the rest of the components of the system hardware. Finally, we proposed a kind of hardware system in micro intelligent vehicle with a hierarchical structure, which was able to provide hardware support for lane detection based on machine vision and traffic sign detection.Secondly, we proposed an optimal lane detection algorithm based on RANSAC spline fitting. For the first, we turned lane images into bird's-eye view by inverse perspective mapping in micro platform and decreased the degree of difficulty in the lane detection. Then, we detected the edge of the lane applying LBP algorithm and Hough transformation. In the end, we adopted RANSAC algorithm to detect the lane combine with spline fitting. So the algorithm could extract the lane in miniature platform effectively and provide a strong guarantee for the autonomous driving of micro intelligent vehicle.Finally, we presented a hierarchical traffic sign detection algorithm based on SVM binary tree. First of all, to detect traffic signs by using color combination and shape characteristics. Secondly, we divided the region of interest into simple categories by determining the main components of color in it. At last, we calculated the HOG in region of interest, recognized the meaning of the interest region by using HOG as identifying characteristics and applying the recognition method of support vector machines based on binary tree. The results of the experiments showed that the algorithm had a higher recognition rate for traffic signs in the micro simulation environment.Through upgrading the micro intelligent vehicle hardware platform and improving the visual perception algorithm, this article ultimately achieved the basic functions of micro intelligent vehicle based on visual perception, including the lane detection and traffic sign detection, which would lay a foundation for the driverless intelligent vehicle in real tests.