Monocular camera-based human body orientation measurement system

The human image data acquired by monocular camera contains rich effective information, and as the basic information of human body orientation, it can be used to infer the intention of the person and the movement situation and other advanced information, so its accurate measurement is of great significance for the gait recognition and other research fields. The current neural network-based methods have the problems of high computational cost and susceptibility to interference. In this paper, based on the law that the body always faces the forward direction when a person walks, we adopt a human orientation measurement method based on trajectory recognition, use the single-response transformation method to remove perspective, combine with the Douglas-Puke thinning and other methods to remove undesirable perturbations of the trajectory, and build a control measurement system with the neural network method. The experimental results show that the measurement method proposed in this paper has a measurement accuracy of up to 91.4%, and at the same time, it has good robustness, especially in the scenarios that require high accuracy of the human body's orientation angle, and improves the accuracy by 8.2% compared with the neural network method. The trajectory recognition-based approach to processing video sequences in real-world scenarios can be used to improve the quality of datasets applicable to areas such as gait recognition and promote the development of this field.