A system for road sign detection, recognition and tracking based on multi-cues hybrid

This paper presents a road signs detection, recognition and tracking system based on multi-cues hybrid. In detection stage, the color and gradient cues are used to segment the interesting regions, and the corner and geometrical cues are used to detect the signs. A pseudo RGB-HSI conversion method without the need of nonlinear transformation is presented for color extraction. In recognition stage, a coarse classification is performed using the corresponding relationship of color and shape, then the Support Vector Machines with Binary Tree Architecture is built to recognize each category of road sign. Furthermore, we present a finite-state machine to decide whether a road sign is really recognized by fusion multi-frame recognition results or not. In order to reduce recognition errors, Lucas-Kanade feature tracker is introduced for road sign tracking. Experimental results in different conditions, including sunny, cloudy, and rainy weather demonstrates that most road signs can be correctly detected and recognized with a high accuracy and a frame rate of approximately 15 frames per second on a standard PC.