Work-related musculoskeletal diseases (WMSD) are a common non-fatal occupational injury. WMSD can lead to prolonged physical pain as well as illness. This results in the worker’s inability to carry out normal work tasks, affecting productivity and even leading to absenteeism. Risk assessment is the primary tool for identifying jobs or operations at risk of WMSD. Risk assessment evaluates the degree of difficulty of an action by observing the elements of the worker’s action, and then integrates the scores of these risk factors with different weights, thus identifying and evaluating the potential risks in the workplace. However, there are still two problems with the current research on risk assessment: limitations in the assessment methods, the complexity of the work scene and the size of the staff. This study proposed Convolutional neural network (CNN) based high-resolution network (HRnet) deep learning network architecture. The method is based on an automated camera detection approach, which is able to efficiently extract low-level spatial information and increase recognition accuracy. The pose detection model is improved by adopting the yolov3 network model and incorporating a target detection algorithm to detect the human body. This improvement enables multi-person pose detection, thereby expanding the model’s range of application, increasing recognition accuracy and making it suitable for multi-person pose estimation tasks. The experimental results show that the proposed model achieves a high level of accuracy. The model supports postural risk assessment in complex work environments as well as multi-person collaborative tasks.