Traffic Lights Recognition Based on Position Feature

Due to the color and shape characteristics of traffic lights, the color model and shape detection are used to detect traffic lights in the work, the images were processed by the ROI (region of interest) extraction, image enhancement, grayscale binarization processing and morphological processing. Then the contour search and connected domain filtering algorithm were used to extract the traffic signal backlight backplane area, thus detecting and segmenting the traffic signal light backplane. Moreover, taking the traffic signal light backplane as positive sample, the other non-traffic light backplane was used as negative sample to build model library. HOG algorithm was used to extract the feature vectors of samples and exclude the false targets based on SVM classification algorithm. Finally, according to the positions of red and green signal lights on signal light, the pixel value accumulation in the area where the signal light is located was calculated as position feature to recognize the red and green signal lights.