Desarrollo e implementación de un sistema de visión artificial para la identificación de posturas ergonómicamente correctas en el área de corte y costura de una MYPE de confección textil en Trujillo – Perú

This project aimed to develop and implement a computer vision system for detecting ergonomically correct postures in the cutting and sewing area of a textile company in Trujillo, Peru. The hypothesis was established that implementing a system based on LSTM neural networks would significantly improve accuracy in detecting incorrect ergonomic postures compared to traditional manual evaluation methods. Advanced image capture techniques and machine learning with LSTM neural networks were used for real-time work posture analysis. The tools employed included Mediapipe for body keypoint detection, and TensorFlow and Keras for data processing and training. The results demonstrated that the system achieved an average accuracy of 93% in identifying correct postures, with a processing speed of 0.5 seconds per image. The solution significantly outperformed traditional manual methods in accuracy (93% compared to 85%), improving ergonomic conditions and reducing occupational risks in the workplace. It is concluded that the software's modularity and scalability ensure its adaptation to future requirements, contributing to a safer and more efficient work environment for the textile company.