Traffic Sign Recognition Algorithm Based on Multi- Modal Representation and Multi-Object Tracking

An algorithm for traffic sign recognition and tracking is proposed in this paper. Image segmentation based on color space conversion and shape classification based on signature feature are used to detect traffic signs in complex urban scenes. To improve recognition accuracy, a two-modal representation method is presented to classify the detected candidate regions for traffic sign. One modal utilizes 2D independent component analysis(2DICA) followed by dualtree complex wavelet transform (DT-CWT), and the nearest neighbor classifier is employed later to classify the traffic sign images and reject the noise regions. The other modal is template matching based on intra pictograms of the traffic signs. The recognition results of the two representations are fused by some decision rules. A multiple-object tracking algorithm is used to track several traffic sign objects in the same scene. The experiment results show that the overall recognition rate of the proposed algorithm is more than 91%, and multiple objects of traffic signs are tracked steadily and effectively. It is proved that the proposed method is robust, effective, and accurate to classify and track the traffic signs.