Pose Detection

gst-ai-pose-detection performs human pose detection on each frame of a video stream, drawing skeleton connections between body keypoints.

Uses the HRNet model.

Prerequisites

• Completed QIM SDK Installation and Model Download

Steps

1. Verify Required Files

radxa@airbox$

ls -l /etc/models/hrnet_pose_quantized.tflite
ls -l /etc/labels/hrnet_pose.json
ls -l /etc/labels/hrnet_settings.json

hrnet_settings.json defines the connections between body joints. Missing this file causes Invalid pose settings path error.

2. View Configuration

radxa@airbox$

cat /etc/configs/config_pose.json

Key fields:

Field	Default	Description
file-path	/etc/media/video.mp4	Input video path
ml-framework	tflite	Inference framework
model	/etc/models/hrnet_pose_quantized.tflite	Model file
labels	/etc/labels/hrnet_pose.json	Label file
pose-settings-path	/etc/labels/hrnet_settings.json	Joint connection config
runtime	dsp	Inference hardware

3. Run

radxa@airbox$

gst-ai-pose-detection --config-file=/etc/configs/config_pose.json

Press Ctrl + C to stop.

Expected Output

Terminal output:

Running app with model: /etc/models/hrnet_pose_quantized.tflite and labels: /etc/labels/hrnet_pose.json and settings /etc/labels/hrnet_settings.json
Using DSP Delegate
VERBOSE: Replacing 518 out of 518 node(s) with delegate (TfLiteQnnDelegate) node
Pipeline state changed from PAUSED to PLAYING

The display shows the test video with skeleton connections drawn between body keypoints.

Validation

• Using DSP Delegate: Inference running on NPU
• Replacing 518 out of 518 node(s): All 518 operators delegated to DSP
• Pipeline reaches PLAYING state
• Display correctly shows human skeleton overlay

Troubleshooting

Invalid pose settings path

Verify hrnet_settings.json exists under /etc/labels/:

radxa@airbox$

ls -l /etc/labels/hrnet_settings.json

If missing, re-run the download script:

radxa@airbox$

sudo ./download_artifacts.sh -v GA1.6-rel -c QCS9075