Frame Rate Optimization in Traffic Sign Detection

Traffic sign plays an important role to ensure maximum transportation safety for people. In an Autonomous Vehicle (AV) system, traffic signs help AV to get more familiar with the road conditions and its surrounding. AV employs Artificial Intelligence (AI) for recognizing its surrounding by detecting and classifying the traffic signs. However, AI has a high computational algorithm which takes a lot of time to finish a detection process and therefore produces a low frame rate which may lead to a fatal accident. Meanwhile, the application of a Graphics Processing Unit (GPU) in AI will increase the system's frame rate but requires an expensive cost. This paper proposes an algorithm to increase the frame rate of a traffic sign detection system without an additional GPU. Our approach is to shift the color and shape segmentation systems right before sending the frames to the main system. Since the number of frames with traffic signs typically less than the ones without traffic signs, we can reduce the computation load of the main system, and therefore a higher frame rate will be achieved. We show that the pre-processing system improves the frame rate of the system to reach 19.4 frames per second.