Human detection under static background may be not suitable to video surveillance which is of complex background, low resolution, small target and noise jamming. All these polytropes and possible occlusion that pose new challenges for us to recognize human after motion detection in single image. For this reason, this article design and implement an human detection algorithm using integral histogram instead of HOG under complex background based on HOG human detection algorithm. Then the cascade SVM is adopted in my experiment to train the sample. Experiment showed that the algorithm has higher precision and faster speed than other algorithm under video surveillance with complex background.