A method for detecting structural surface cracks based on image features, support vector machines and Bayesian data fusion, including: 1) acquisition of a video of a structural surface and establishment of an image library; 2) calculation of texture features of the image frames by local binary patterns; 3) scanning and grouping for image patches of cracks on the image frames using two-stage support vector machine; 4) Bayesian data fusion and decision. The video image detection acquires images of many areas where human beings are difficult to reach; computers are adopted to identify cracks on surfaces of the structural elements, which can greatly reduce the identification workload and labor cost, and increase the crack detection rate. The invention has a better adaptability to the light strength on the structural surface, thus providing better identification for cracks.