Research On Methods Of Traffic Lights Detection And Recognition In Complex Scenes

With the increasing popularity of automotive vehicles, traffic problems such as road safety and traffic efficiency are becoming increasingly prominent. In order to effectively reduce the probability of the occurrence of traffic accidents, road traffic lights recognition, which is based on computer vision and pattern recognition, is becoming an important research area of Intelligent Transportation System and one of the key techniques for the design of intelligent vehicles. The core technology in traffic lights recognition is in algorithm. However, although computer technology and artificial intelligence technology develops rapidly at present, as well as target detection and recognition algorithms are emerging continuously, there is no existing algorithm can handle the traffic light recognition problem well duo to the variety of ambient light, the complexity of environment, the real-time requirement and so on. And the existing traffic lights recognition research is limited to the motor vehicle (round) lights in simple scenes. To solve the above problems, this paper presents a method of traffic lights detection and recognition in complex scenes. The method can both satisfy the needs of recognition for traffic lights in complex city environments and recognition for directional signal lights. This paper makes the following researches:On the basis of fully tap the traffic lights characteristics; the paper presents the system framework for traffic lights detection and recognition, which is composed of detection, recognition and tracking three components. Secondly, to satisfy the need of computer vision applications of intelligent vehicles, the paper proposes an image acquisition and storage system. The system solve the problems of image acquisition and its storage interference, high resolution and high frame rate mutually exclusive in image acquisition, multi-image interference in image transmission, and storage problems with IO operation and compression algorithm.Traffic lights detection is the base of traffic lights recognition. The paper uses color segmentation and associated filter to implement traffic lights recognition. First the Gaussian model of traffic lights is established and the paper proposed image segmentation using Gaussian vector and multi color spaces. Then the paper proposed the associated filter applying region growing algorithm and similarity comparing to filter the image after segmentation. Complex scenario experiments show that this method can improve the correctness of the detection of traffic lights, reducing the false alarm rate and missing rate.To solve problems of existing traffic lights recognition algorithms, which can't applied to recognition for directional traffic lights. The paper presents a method using canny operator to extract the edge at first, then using revised hu invariant moments and Mahalanobis distance to classify the directional signal lights. Experiments show that this method can meet the directional signal lights recognition requirements and can be extended to recognize signs with a specific outline.