Abstarct: One of the main applications of intelligent video surveillance is in machine vision. The general configuration of these systems is a set of cameras with a non-overlapping field of view. The ultimate goal of these systems is to detect abnormal events and analyze individual behaviors. Human re-identification is one of the main steps in the system, which is to label people in front of a network of cameras, so that a single label is assigned to a particular person. In such systems, a person can be placed in front of a set of cameras at different times and positions. For various reasons, recognition is one of the complex problems of computer vision. The reason for this complexity is the change in a person’s appearance from one camera to another. These changes can be caused by hardware limitations such as low-end cameras, different camera responses to colors, changing optical conditions in different camera positions or angular variations in how people are exposed to cameras. One of the main weaknesses in the existing methods is the lack of attention towards angular information. In fact, when the quality of recorded images is low or people are far from the camera, existing algorithms inevitably ignore important information, such as angular information. The purpose of this study is to model the judgment of the human supervisor regarding visual changes in the coverage of the target subject, resulting in a change in the angle of view. For this purpose, the orientation of the body towards the camera and the resulting 3D changes are used. In the proposed approach, the orientation of the person in the image is first extracted by two proposed methods, one baxsed on contour information, and the other on the histogram of the oriented gradients and the logistic regression classifier. Then, certain areas that can be seen or hidden under angular rotations are identified. To this end, the region of the head in the image is used as a key point that is extracted using a convolutional neural network. In the matching stage of the re-identification process, the extracted areas are weighed, which we call it unification process in the proposed approach. In designing the proposed system, we have considered the computational complexity in all parts. The significant feature of this system is the ability to use and integrate in most re-identification methods. In this research, the proposed approach was integrated into two pioneering algorithms in the field of re-identification. The results of applying this method to the ViPER dataset, which is considered as one of the most difficult re-identification datasets, indicate an improvement of 1.3% in the recognition rate with respect to the two aforementioned algorithms and the effectiveness of the proposed unification process.