概览
描述
Action Recognition with pose estimation solution is developed and inferenced using RZ/V series vision MPUs with DRP-AI acceleration. This solution recognizes a person's actions with body key points from HRNet inferenced on DRP-AI. Actions were defined with certain body actions such as fallen, panic expressions, sick, eating, etc. This application can be inferenced using the RZ/V2MA vision board.
特性
- Pretrained HRnet with 80% accuracy for generating key human body points
- Actions defined with training and test method for body key points
- Action recognition with body points system inferenced with 90% accuracy
- Comprehension of HRnet pose estimation key points and devise an algorithm capable of detecting actions such as falling, coughing, eating, and vomiting
- Application for performing action recognition on an Edge AI accelerator, enabling inference of the algorithm and conducting post-processing to display outputs
- Inference successfully performed on an Edge AI accelerator with a fast inference time of 50 ms at a speed of 10 FPS for defined actions