The Lane Line Detection And Traffic Sign Recognition Base On Computer Vision

As technology and unmanned vehicle auxiliary system get the appreciation of more and more automobile enterprise, a series of research on the background of the driverless technology get more and more the attention of the academia and industry. Several well-known vehicle companies, such as Volvo, Infiniti, Mercedes, and Hongqi have used some mature vehicle auxiliary system technologies such as adaptive cruise system, "active lane control system" and "active steering by wire" technology in the design of their products. Google is the first non-car companies in study of unmanned technology, its design of the unmanned vehicle has passed the 300000- mile test. Domestic unmanned technology started late, but it developed rapidly, unmanned car invent by the national university of defense technology has passed from Changsha to Wuhan section on high speed road. Unmanned technology involves many fields: automatic control, architecture, artificial intelligence, computer vision.Research suggests that the key point and difficulty of unmanned technology research is to analysis of the surrounding environment of vehicle, and computer vision is used to solve this part is just right. This article mainly studied from two aspects of unmanned technology based on computer vision: the lane line detection and traffic sign recognition. For the first part, this paper proposes an algorithm: An empirical model which based on HSV space and connected field. This algorithm can solve three exist problems: First, algorithm only can detect straight line. Second, algorithm can’t distinguish full line and dash line. Third, algorithm can’t solve the problem with multi-line detect. For the second part of this paper designed a traffic sign recognition system combine detection with recognition, the system combine the color information and shape information for detecting, HOG + SVM classification algorithm for classifying. Training classification model use 23429 traffic sign samples, and then use the classification model to recognize road signs, its classification accuracy reach 88.9725%. The system is able to detect and identify 20 categories of traffic sign(prohibit and indicative), and can be present the final recognition results to the interface designed.