三维人体关键点检测方法、装置、电子设备及存储介质
By using a separate 3D human keypoint prediction model and generating pixel coordinate and depth coordinate heatmaps through convolutional neural networks, the problems of speed and accuracy of 3D human keypoint detection are solved, making it suitable for mobile terminals.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JINGDONG TECH HLDG CO LTD
- Filing Date
- 2023-01-09
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, three-dimensional human key point detection methods have the problem of balancing speed and accuracy. Three-dimensional heat map prediction models have many parameters and large computational load, making them unsuitable for mobile terminals. Regression prediction models have low prediction accuracy.
A separate 3D human key point prediction model is adopted. Image features are extracted through convolutional neural networks to generate pixel coordinates and depth coordinate heatmaps of human key points, which are then fused to obtain 3D coordinates. The training process and the inference process are separated, which reduces the amount of computation and improves accuracy.
It achieves fast and accurate 3D human key point detection, reduces the number of model parameters and computational load, and is suitable for mobile terminals.
Smart Images
Figure CN115965996B_ABST