A Review on Action Recognition and Its Development Direction

Human action recognition has gradually developed into a popular research field. This article starts from the steps of action recognition and the methods commonly used in each step. It divides it into two parts: motion tracking and analysis, and definition of motion. In the first step, three commonly used methods based on bone joint points, based on optical flow and based on points of time and space are introduced, and their advantages and disadvantages are further elaborated. The commonly used method in the second step is based on neural networks and graph models. In this paper, CNN, RNN, LSTM and other mainstream algorithms are used to introduce neural networks. Hidden Markov models and restricted Boltzmann machines introduce graph models. Finally, the development direction of behavior recognition is discussed, and multi-person interactive behavior recognition and behavior recognition in other fields, such as vehicle behavior recognition, are proposed.