Hand Tracking Libraries #2
Description
OpenPose
OpenPose requires a discrete GPU for real-time throughput (typically 30+ FPS at 368×368 resolution on a modern NVIDIA card) and falls back to slower CPU‐only processing if no GPU is available.
https://github.com/CMU-Perceptual-Computing-Lab/openpose
IntelRealSense
This comes from RealSense itself but the software seems to be abandoned and experimental.
https://github.com/IntelRealSense/hand_tracking_samples
Google's Media Pipe
Compute requirements can be nontrivial on CPU-only hardware. While the pipeline can achieve 30–60 FPS on mobile GPUs or desktop cards, on a CPU (or low-power embedded board without an accelerator) you may see single-digit frame rates unless you aggressively downscale inputs or reduce model resolution
https://github.com/google-ai-edge/mediapipe/blob/master/docs/solutions/hands.md
Can probably use this
These seem to be already written APIs over the mediapipe models?
https://github.com/Vibhu04/mediapipe_hand_tracking_cpp
https://github.com/Mario-td/Simplified-hand-tracking-with-Mediapipe-CPP/tree/main
https://www.analyticsvidhya.com/blog/2021/07/building-a-hand-tracking-system-using-opencv/
This just uses OpenCV
https://github.com/PierfrancescoSoffritti/handy/tree/master/Handy