BlazePose: On-device Real-time Body Pose Tracking
推荐理由
轻量级二维人体姿态估计经典方法，具有算力资源要求低、准确度高等特点，可部署在笔记本电脑、移动手机等边缘设备上，当下被广泛应用于使用轻量级二维人体姿态估计方法实现各类运动分析的应用中。
| 文章简介 | |
|---|---|
| 期刊 | CVPR Workshop on Computer Vision for Augmented and Virtual Reality, Seattle, WA, USA, 2020 |
| 发表年份 | 2020 |
| DOI | 10.48550/arXiv.2006.10204 |
| 类型 | 研究性工作 |
| 领域 | 计算机视觉 |
| 引用量 | 467 |
| 推荐信息 | |
|---|---|
| 推荐人 | 阿璐思 |
| 审核 | 王伟明 张琛 |
| 推荐小组 | 动捕小组 |
摘要
We present BlazePose, a lightweight convolutional neu- ral network architecture for human pose estimation that is tailored for real-time inference on mobile devices. During inference, the network produces 33 body keypoints for a sin- gle person and runs at over 30 frames per second on a Pixel 2 phone. This makes it particularly suited to real-time use cases like fitness tracking and sign language recognition. Our main contributions include a novel body pose track- ing solution and a lightweight body pose estimation neural network that uses both heatmaps and regression to keypoint coordinates.
细分领域