Vehicle Taillight Detection And Semantic Recognition

The lamp semantics of driving is a major way in which vehicles transmit information in traffic flow each other. It is of great significance and broad applications to realize the detection and recognition of taillight semantic automatically. However, it is a challenge to detect and recognize the taillight semantic from an image or video under real urban road environment condition in daytime. This paper research on the above problem. The summary of the main work as follows:1:A new detection method of taillight is put forward, which can be applied to a real road conditions.Firstly, the threshold of color segmentation is determined based on the statistics of taillights color distribution in the road environment during daytime. Using the HSV color model to extract the area, and set up the area with different weighted, then, nonlinear transform of RGB component enhanced images are carried out using these different weighted coefficient. Thus, good segmentation results is obtained. Then, according to the symmetry characteristics of taillight, pairs of taillights matching validation method is proposed by using the position and area of constraints. Finally, we tracked taillights by using the algorithm of Kalman filter. The same taillights have different locations in different frame images, but which are correlative for the frames are continuous. Using the correlative information, then we achieved an effective and stable detection of the taillights.2:A new algorithm of taillight semantic recognition is proposed. Firstly, the luminous mode of brake lights and turn lights are analyzed, and the characteristics of the brake signals, turn signals and lights turned off are studied. Secondly, the method of using two levels classifier to recognized semantics of taillight based on Support Vector Machine (SVM) is proposed. According to the classification task of each level classifier, the different classification features is design for training and classification then, different design methods of classification features are discussed and experimented. Finally, combining the classification results with the frequency of turn lights’illumination, the final recognition results is obtained.At last, the proposed methods in this paper are carried out and discussed. Experiment in actual road image show that the accuracy of taillight detection and taillight semantic recognition is94.6%and79.8%respectively...