Kinect Sensor and Improved ResNet-34 Model for the Design of a Precision Training Information Sharing Platform for Vocational Skills

The development of technologies such as computers and artificial intelligence provides excellent technical support for the construction of precise vocational skills training platforms. In order to facilitate the sharing of vocational skills training information and improve the efficiency of workers, the study adopts the Kinect depth sensor to capture the vocational skills posture information of the trainers, and the improved deep residual neural network-34 to recognize the posture information of the captured data. The study introduces multi-scale feature fusion and transfer learning techniques to optimize the deep residual neural network-34. In order to build the final information sharing platform, the study utilizes a browser/server architecture. The results show that the maximum accuracy of the vocational skills pose recognition model for training teachers constructed by the study is 99.24%, 99.87%, and 99.73% with the color, depth, and skeleton data acquired by Kinect, respectively. The maximum response time of the information sharing platform is 1484 ms. When the amount of shared data is 1000 MB, the central processor occupancy, transmission efficiency and central processor transmission efficiency of the platform are 3.21%, 18.1549 Mbps and 20.3983, respectively. The pose recognition model and the information sharing platform constructed by the study have good performance and can provide platform support for the sharing of vocational skill accurate. It can promote the improvement of labor efficiency.