An In-Depth Study and Analysis of Human Gesture Action Recognition Based on Deep Learning Foundation

In the traditional human posture estimation algorithm, the convolution kernel for feature extraction of the input feature map almost always uses the standard convolution kernel, in order to ensure the accuracy of the model, this convolution kernel inevitably needs to consume a large amount of memory, and the process of human posture estimation often involves the detection of multiple key points of the human body, which is often a very high demand on the algorithm, which will lead to a large amount of wasted computational resources. With the development of science and technology, smart devices are becoming more and more popular in our daily life, but due to the fact that the number of parameters and computation volume of human posture estimation are often very large, it is basically impossible to apply this kind of network model of high complexity to our daily life only devices. Therefore, it is necessary to study how to make the network model lighter while ensuring the performance, so that the human posture estimation technology can be applied to mobile devices. In this paper, the algorithmic foundation of deep learning is based on the human posture action recognition in-depth research and analysis, which gives a new research method and research technology for human action recognition, and the research in this paper promotes the application of deep learning and the research progress of human posture action recognition.