Smart Motion Detection with Raspberry Pi AI Camera

Get the Sony AITrios sample apps repository and set up the Python environment.

1. Open Terminal on your Raspberry Pi.
2. Clone the repository with the motion sensor code.
3. The uv package manager will automatically install dependencies from pyproject.toml when you run the app.

$ git clone https://github.com/SonySemiconductorSolutions/aitrios-rpi-sample-apps.git
cd aitrios-rpi-sample-apps/examples/motion-sensor

Set up polygon regions where motion should trigger capture.

1. The example.json file defines Areas as polygon coordinates. These restrict detection to specific regions like a doorway or loading zone.
2. You can edit example.json manually to change the monitored area coordinates, or use the in-app configurator tool later.
3. For now, the default example.json works to test the entire camera view.

Start the motion-triggered image capture app and see it detect moving objects.

1. Run the app with uv, which installs dependencies automatically and starts the detection pipeline.
2. The app uses NanoDet for object detection, BYTETracker for stable IDs, and a Motion module for pixel-change analysis.
3. When a tracked object moves (and optionally stays inside your defined Area), the app saves a timestamped image to ./images.

$ uv run app.py --json-file example.json

Learn the six-component pipeline that makes smart motion detection work.

1. The system runs six stages: (1) NanoDet object detection produces bounding boxes and class IDs. (2) Detection filtering applies confidence thresholds and optional class filters (e.g., only detect 'person'). (3) BYTETracker assigns persistent IDs so the same object keeps the same ID across frames. (4) Motion estimation computes motion bounding boxes from pixel changes. (5) Area gating restricts checks to user-defined polygon regions. (6) Matching logic confirms a detected object is also moving, then triggers image capture.
2. This merge of AI detection and pixel motion avoids false triggers from lighting changes or static objects, and reduces storage by capturing only when it matters.

Record video clips instead of single images when motion is detected.

1. Stop the image-capture app if it's still running (press Ctrl+C).
2. Run the alternate app_video.py script, which records video clips instead of saving single frames.
3. Recorded clips are saved with timestamps, making it easy to review events.

$ uv run app_video.py --json-file example.json

Tune confidence thresholds, class filters, and trigger logic for your use case.

1. Open app.py in a text editor to adjust settings.
2. Change the confidence threshold (default 0.5) to make detection more or less sensitive.
3. Filter to specific classes (e.g., only trigger on 'person' detections) by adjusting the class_id filter.
4. Modify the trigger logic in the check() function to change when captures happen (e.g., require motion for more frames before triggering).