E2Pose: Fully Convolutional Networks for End-to-End Multi-Person Pose Estimation
Masakazu Tobeta, Yoshihide Sawada, Ze Zheng, Sawa Takamuku, Naotake Natori · 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · 2022
Highly accurate multi-person pose estimation at a high framerate is a fundamental problem in autonomous driving. Solving the problem could aid in preventing pedestrian-car accidents. The present study tackles this problem by proposing a new model composed of a feature pyramid and an original head to a general backbone. The original head is built using lightweight CNNs and directly estimates multi-person pose coordinates. This configuration avoids the complex post-processing and two-stage estimation adopted by other models and allows for a lightweight model. Our model can be trained end-to-end and performed in real-time on a resource-limited platform (low-cost edge device) during inference. Experimental results using the COCO and CrowdPose datasets showed that our model can achieve a higher framerate (approx. 20 frames/sec with NVIDIA Jetson AGX Xavier) than other state-of-the-art models while maintaining sufficient accuracy for practical use.