A new Raspberry Pi AI camera brings computer vision to all Raspberry Pi models

Earlier this year, Raspberry Pi launched its AI Kit which supports more complex artificial intelligence workloads than those natively supported by its strongest-performing hardware. However, the AI Kit is only compatible with the Raspberry Pi 5, unlike the new AI Camera module which is compatible with all Raspberry Pi models. The AI Camera module features a Sony 12 MP Sony IMX500 Intelligent Vision Sensor and an integrated RP2040 AI accelerator.

The Sony image sensor has a 78-degree field of view with manually adjustable focus, and two operative modes: 4056×3040 at 10fps, and 2028×1520 at 30fps. The integrated RP2040 accelerator handles the sensor firmware and neural networks, leaving the host Raspberry Pi's main processor free to perform other tasks. The whole module can be connected to any Raspberry Pi using a standard camera ribbon cable, and the system can be configured using the Sony AI tools suite to run existing neural network models using frameworks like PyTorch or TensorFlow. It is also possible to design new models that can leverage the specifications of the camera module.

The AI Camera integrates seamlessly with Raspberry Pi's camera software stack, simplifying application-building. The AI Camera introduction announcement showcases two of these in action: an object detection neural network model (MobileNet SSD) running under rpicam-apps, and a pose estimation neural network model (PoseNet) performing inference on a 1080p video at 30fps. To make application building even more accessible, Sony and Raspberry Pi have collaborated to optimize several neural network models, which are now available in Raspberry Pi's model zoo along with some visualization example scripts using Picamera2. Although the AI Camara is more affordable because it includes all the components needed for computer vision in a single module, its only downside is that the accelerator may not be as capable as the one in the AI Kit.