Advantages of openpose human posture recognition

With the development of computer vision and its wide application in daily life, the project of behavior detection and motion recognition based on various algorithms has been applied more and more in practice and has been widely studied in related fields. In the aspect of behavior monitoring, it is not only information such as graphics, temperature and humidity, sound, etc. It is used to monitor the behavior of bees, but more applications focus on monitoring human behavior. As an important reference of behavior monitoring, human posture recognition has been widely used in video capture, computer graphics and other fields. Among them, the traditional human posture recognition methods include RMPE model and mask R-CNN model, both of which are top-down detection methods. As a classic project of attitude recognition, Openpose is a bottom-up detection method, which is mainly used for behavior monitoring, attitude correction and action classification, and has important research significance and application value in smart home, autonomous driving, intelligent monitoring and other fields.

In the aspect of gesture recognition of multi-person target, the commonly used methods in history include searching candidate key points from top to bottom, matching people with spatial connection optimization algorithm, and detecting human skeleton connection from key points by establishing partial affinity field.

Aiming at the low accuracy and efficiency of current behavior monitoring, this project combines openpose gesture recognition technology and constructs a classification algorithm through coordination between different limbs. By comparing different classification algorithms, the optimal model is selected to construct a multi-target classification method, and finally the real-time display of gesture display, target detection and multi-target classification is realized. In this model, the lightweight openpose model is called to recognize human posture. The main method is to get the key points of human bones through openpose, and then match two bones through Euclidean distance to detect everyone. For the missing key points in common detection, the skeleton information of the previous frame can be used to fill them.