With the development of society, the world's population mobility, mobility speed and mobility area are increasing, and face detection and recognition technology also plays an important role in the management and statistics of mobile population, and has great application prospects in digital entertainment. The aim of this paper is to study face detection based on cascaded support vector machines. The performance metrics specified in the training are reduced, effectively simplifying the structure of the cascading classifier, which is then used as a filter, followed by the introduction of a nonlinear support vector machine main classifier based on rectangular features. Experimental results show that although the detection speed does not reach the real-time level of Viola's method, it is still a significant improvement over the pixel-based method, along with a higher face detection rate.