Visual perception of traffic sign for autonomous vehicle using k-nearest cluster neighbor classifier

Recently, developing an autonomous navigating vehicle becomes more attention. It is equipped by multiple sensors, such as radar, laser, GPS, and camera. Among these sensors, utilization vision technique is the most adopted method for constructing such a system. It is because the camera provides a lot of information and is low-cost device rather than other sensors. Traffic signs, as one of the most important visual information, carry a lot of useful information required for navigating. Thus, in this work, traffic sign detection and recognition framework is addressed. First, an input image is converted into normalized red-blue color space, as traffic signs usually appear with red and blue color. Second, maximally extremal stable region (MSER) method is then performed for extracting candidate regions. Using geometry properties, the false regions will be excluded. Third, histogram of oriented gradient method is applied in order to extract features from candidate regions. Lastly, k-nearest cluster neighbor classifier is then processed to classify region into a certain traffic sign class. In experiments, our system achieves 98.07%, 99.54% and 25fps for detection rate, recognition rate and average frame rate. It is almost forty times faster than classical k-NN. These results demonstrate the effectiveness of our systems that can be implemented well on real-time application.