Design And Implementation Of Traffic Lights Detection And Recognition System Based On Visual

With fast economic development, the vehicle has been going into every family’s life, and is more popular. At the same time, traffic conditions are becoming more complex, which brings traffic congestion, frequent accidents and poor road transport efficiency. So, driving safety, traffic efficiency of intersection and other issues cause more and more concern. Traffic lights detection and recognition system (TLS), which is an important part of intelligent transportation system (ITS), can be applied to advanced driver assistance system and unmanned systems.It is difficult to carry out the detection and recognition of traffic lights in the real road environment, because there are many affecting factors such as road conditions, background environment, light, weather and so on. The requirements of real-time and accuracy make this more difficult. The existing detection of traffic lights is mainly for vehicle lights, while study about traffic lights indicating the direction is rarely. Therefore, based on the full study of the characteristic properties of traffic lights, this paper designs a real-time detection and recognition system about both circle and arrow-shaped traffic lights. First, the paper establishes a framework for detection and recognition system of traffic lights. The structure of the system including, preprocessing, detection, recognition, tracking and showing. The system is simple, real-time and easy to be implemented. The following is the main contents.First, doing image preprocessing including brightness filter and color segmentation. In this procedure, color and brightness characteristics of light-emitting unit are used. Then filtering based on the shape of the light-emitting unit is done. After this, it matches and verifies the external rectangle using the back black board of traffic lights. Then the paper does statistical verification of the results of consecutive frame images. Finally, precisely positioning the traffic signal lights is finished and status information is get. Besides, the type of traffic signal lights is identified by feature extraction and matching. Considering the short comings of the current detection and recognition algorithms for traffic signal lights, this paper attempts to do the following points.1. In the detection process of traffic lights, In order to achieve the purpose of reducing erroneous detection, the paper not only uses a black light rectangular housing for validation, but also uses multi-frame statistics for validation which further confirms the traffic lights region.2. Currently there is no accurately geometric model to describe the characteristics of the arrow-shaped traffic lights, so this paper attempts to model arrow-shaped traffic lights by adopting the boundary characteristic points and saturation information.3. To recognize the type of traffic lights, the paper combines boundary feature points and characteristic triangle saturation. Besides, the paper takes Hu feature extraction algorithms to extract boundary characteristics of candidate region for further classifying the traffic lights indicating the direction.After a large number of experimental analyses for the traffic signal detection and recognition system, Experimental results show that the detection and recognition algorithm meets the real-time requirements and has a high recognition rate.