신호등의 형태 특성을 이용한 신호등 탐지 및 인식 알고리즘

Abstract

In this paper, we use a Haar-like Feature algorithm proposed by Viola and Jones for shape-based traffic light detection. Haar-like Feature maintains the geometric information of an object and used the difference of brightness of the area, so it can maintain the shape change and position change of the object inside the area. Also, Haar-like Feature can detect the object in the image very quickly. To improve the accuracy of the traffic light candidate image obtained through the Haar-like feature, the reliability is improved through the verification process with the SVM Classifier. The image processing such as binarization, morphology, and contour detection is performed on the traffic light image obtained through the verification process so that only the accurate traffic light object is extracted. The extracted traffic light objects are classified into three color traffic light or four color traffic light according to the aspect ratio. The divided traffic lights are divided into each signal areas to recognize whether the signal are turned on. It is possible to recognize a signal of a traffic light by recognizing which region of the signal has been turned on.