Raspberry Pi AI Camera Detects People, Triggers OpenPLC Alerts

Sony Semiconductor Solutions published the aitrios-rpi-sandbox repository with sample applications that run on a Raspberry Pi AI Camera and integrate detections with OpenPLC to trigger real-time stack light and LED notifications. The repo demonstrates on-device object detection, pose estimation, and privacy-preserving pipelines that avoid cloud streaming while providing alerts, screenshots, and annotated data capture.

The sandbox bundles Dockerized examples including a Dockerized Flask streaming app and multiple demo flows. Key capabilities shown include:

• •Real-time object detection using the Raspberry Pi AI Camera with local inference
• •Pose estimation workflows using models such as HigherHRNet for keypoint-based logic
• •Privacy masking and neutral-background pose visualizations to reduce identifiable video data
• •Integration points that save images and COCO-style annotations for offline evaluation

The repo also surfaces a model evaluation tool for classification, object detection, segmentation, and pose estimation, enabling COCO mAP and accuracy checks against injected camera data or model interpreter services (Keras/ONNX). Integration with OpenPLC is implemented to map detection events to PLC outputs, which can drive stack lights, audible alarms, or messaging workflows.

Edge-first person detection tied to PLC control targets two practical domains: industrial safety automation and privacy-sensitive monitoring. By keeping inference on-device, the repo reduces network bandwidth, latency, and regulatory exposure from raw video transfer. The combined stack is a useful reference for teams building industrial IoT systems, factory-floor safety mechanisms, or private site monitoring where cloud transit is undesirable.

Practitioners should evaluate model size, latency, and false positive rates for their target Pi hardware revision, and consider PLC safety and fail-safe design for production use. Expect this repo to serve as a rapid prototype path; moving to production will require hardened model validation, tamper-resistant hardware, and integration with facility safety protocols.