Pedestrian pose estimation is our perception technology in autonomous driving. This technology is the integration of functional neural networks with specific perception information in order to improve safety on roads in reality. The algorithm architecture is based on AI human pose and object detection modules with RGB images for multiple messages of human skeleton into sequential deep learning models. This system is capable of model generalization and computation effectiveness in real scene applications.

1.Interface: CAN 2.0 A
2.Image: 1080 x 720 pixel
3.CAN bus: 500Kbps
4.Inference time: Latency≦2 seconds
5.Accuracy: ≧92% at distance≦10公尺

Autonomous driving, Smart city management, Security system.

Pedestrian safety, Security and protection system in city