Light signal recognition is one of the key issues in autonomous vehicle. Unlike normal object detection and recognition, which can be done by using different sensors, light signal recognition is naturally a computer vision problem. Although commercialized ADAS (Advanced Driving Assistance System) products, such as mobileye, could be used for rear-end collision warning, a cost-effective approach is still needed. In this paper, we have developed a novel two-stage approach to detect vehicles and recognize signal lights from a single image in real-time. Distinct from current state-of-the-art light recognition algorithms, we adopted a HDR (High Dynamic Range) camera instead of a color camera. Taking advantage of the different dynamic range of a HDR camera, the detection of a vehicle and the recognition of the signal light have been done in bright and dark channels, respectively. Furthermore, unlike previous approaches where pair taillight has to be extracted explicitly, we use detected vehicle region instead. On a large database, “Lights Patterns” (LPs) are learned by a multi-layer perception neural network. The robustness and efficiency of the proposed approach has been verified by the experimental results conducted on some real on-road videos.