The accurate traffic light detection is an important role of automated vehicle driving. This paper focuses on a traffic light recognition algorithm by using computer vision and machine learning. In order to improve accuracy and reduce false detections, Region-Of-Interests generated from GPS and predefined map information is useful to identify target traffic signals. Therefore, we have used traffic signal database based on a precise point cloud map. We have evaluated the performance of typical detection methods such as AdaBoost, template matching and deep learning for driving images on public road. The results show the reasonable performance within practical computational time.