System design for preventing work accident behaviors based on artificial intelligence image processing in industrial machines

In this study, a design is being developed using image processing techniques with artificial intelligence to prevent hand injuries and limb losses due to work accidents in workers working on circular saws and similar dangerous machines, and to teach them correct working behaviors. Computer vision is performed using the MediaPipe Hands machine learning model developed with CNN (convolutional neural network) and similar deep learning techniques. The design is implemented in Python using the TensorFlow machine learning library and the OpenCV image processing library. The system uses live camera images to monitor the human hand's proximity and fast movements in dangerous areas. Based on these measurements, danger and warning zones are identified. If the hand approaches too closely to the cutter or shows rapid movements, the machine is stopped. In the warning zone, necessary alerts are provided to the user to avoid accidents. The performance of the MediaPipe Hands model is evaluated under different brightness and noise conditions. To reduce performance loss caused by environmental factors, filters are applied. The brightness compensation and noise-cancelling median filter provide significant improvements in varying lighting and noise levels. Additionally, the model's complexity and its real-time operation capabilities are assessed. The system is expected to reduce work accidents significantly, depending on advancements in cameras, computers, and software.