To detect forward obstacles for intelligent vehicle in complex taffic scenes, a novel obstacle detection method based on wavelet transform module maximum (WTMM) and support vector machine (SVM) was presented, considering obvious rear visual features of forward obstacles. Firstly, the candidate regions of obstacle (ROIs) were detected based on multi-scale singularity analysis with WTMM and multi-features combination of obstacle knowledge. Then, these ROIs were recognised based on a compatible binary tree support vector machine (BT-SVM) classifier for obstacle pattern of traffic scenes. Applied the proposed method to different traffic scenes (e.g., simply structured highway, complex urban street), the online experiment results show the efficient, real-time and universale ability.
