Abstract: An obstacle detection method based on wavelet transform module maximum(WTMM) and ensemble learning support vector machine(SVM) is proposed for outdoor mobile robot in complex traffic scenes.Firstly,the WTMM-based singularity analysis theory is introduced into the regions of interest(ROIs) detection of candidate obstacles.Then,an improved multi-class SVM based on ensemble learning is constructed to classify and recognize these ROIs.In the experiment, the proposed method is applied to various traffic scenes(e.g.,simple highway,complex urban street),and the result proves the validity,universality and real-time performance of the proposed method.