The invention discloses a design method of a multi-pose human face detector based on MSNRD features, including a training phase and a detection phase; in the training phase, the size of the template is first determined, and a training data set is prepared; secondly, a feature pool is constructed; and then in the Adaboost framework Next, train a strong classifier; in the detection stage, first traverse the sub-images of each scale in each position of the input image to determine whether it is a human face; all face images form a set, and the overlapping area of any two elements in the set accounts for the smallest element area If the ratio exceeds 0.3, only keep the element with the larger output value of the classifier, and repeat this step until the ratio of the coincident area of any two elements in the set to the area of the smaller one does not exceed 0.3, and the sum of the set is regarded as the final result of the detection The result output. The method of the invention enriches the expression ability of the feature, simplifies the calculation process of the feature, avoids separately training detectors for different attitudes, reduces the workload of training, and improves the detection efficiency.