Value evaluation of human motion simulation based on speech recognition control

With the popularity of sports wearable smart devices, it is no longer difficult to obtain human movement data. A series of running fitness software came into being, leading the nation's running wave and greatly promoting the rapid development of the sports industry. However, a large amount of sports data has not been deeply mined, resulting in a huge waste of its value. In order to make the data collected by smart sports equipment better serve the sports enthusiasts, thereby more effectively improving the degree of informatization of the sports industry, this paper selects the design and implementation of the human motion recognition information processing system as the main research content. This article combs the previous research results of human motion recognition information processing systems related to sports wearable intelligence and proposes a three-layer human motion recognition information processing system architecture, including data collection layer, data calculation layer, and data application layer. In the data calculation layer, different from the traditional classification algorithm, this paper proposes a classifier based on the recurrent neural network algorithm. The mechanical motion capture method mainly uses mechanical devices to track and measure motion. A typical system consists of multiple joints and rigid links. Inertial measurement units are bound to the joints to obtain angles and accelerations, and then analyze the human body motion based on these angles and accelerations. From the perspective of optical motion capture, the Kinect somatosensory camera is researched, and the method of human motion capture based on depth images, and the principle and method of human motion information are analyzed. At the same time, research on the application of Kinect's motion capture data. As a deep learning algorithm, convolutional action recognition model has the characteristics of being good at processing long and interrelated data and automatically learning features in the data. It solves the defect that the traditional recognition method needs to manually extract the motion features from the data, the whole system structure is streamlined, and the recognition efficiency is higher. The overall evaluation is as high as 99.4%. It avoids the manual extraction of time-domain and frequency-domain features of time series data, and at the same time avoids the loss of data information caused by dimensionality reduction.