The Luxonis OAK-D Pro2 represents a new generation of spatial AI cameras that combine stereo depth sensing with on-device neural network inference. Built around the Intel Myriad X VPU, it runs AI models directly on the camera — no host GPU required.
Architecture
- Depth Engine: Stereo depth with onboard disparity computation, providing dense depth maps at up to 120 FPS
- Neural Inference: Intel Myriad X VPU runs trained models (YOLO, MobileNet, custom networks) at the edge
- RGB Camera: 12MP color sensor for object detection and scene understanding
- IMU: Integrated BNO086 for orientation and motion tracking
- Interface: USB-C for power and data; also available in PoE variants
What Sets OAK-D Pro2 Apart
Unlike traditional depth cameras that only provide raw depth data, the OAK-D Pro2 processes AI inference on-device:
- Object detection results are available with near-zero additional latency
- Host CPU is freed from inference workload
- Bandwidth requirements are reduced — only processed results need transmission
- Privacy-sensitive applications benefit from on-device processing
Use Cases
The OAK-D Pro2 is finding adoption in:
- Autonomous drones with onboard obstacle detection
- Industrial quality inspection with AI defect detection
- Smart access control with facial recognition and depth anti-spoofing
- Robot arm guidance with spatial object detection
Leave a Reply