A human body reidentification method based on portion segmentation is disclosed. The method comprises steps of image data acquisition, rough identification, fine identification and the like. The method has effects that modern clothes have many styles and colors, a color of sleeves is different from a color of a body portion, and colors of trousers of a thigh portion and a crus portion are different; in a monitoring camera, head information is usually subtle so that a characteristic with a large distinction degree is not easy to find, and feet are small and are not easy to collect; and based on the above condition, a human body, except for the head, is divided into 9 portions of a body, a left/right upper arm, a left/right lower arm, a left/right thigh and a left/right crus, and reidentification is performed and a high identification rate is possessed. Through rough identification, a candidate set human body whose color is close to that of a query set can be extracted, but a search scope can not be reduced. Based on the rough identification, for a body portion image, a method of projection is used to extract a pattern of the body portion and position information of the pattern so that the identification rate is further increased.