Raspberry Pi has introduced a new AI Camera priced at $70, enhancing its range of vision-based products with built-in artificial intelligence capabilities. Equipped with Sony's IMX500 Intelligent Vision Sensor, the camera integrates an AI accelerator directly into the image sensor, offering developers an efficient way to run neural network models without heavily taxing the Raspberry Pi's processor.
AI meets Raspberry Pi: what's inside the new camera
The AI Camera boasts a 12-megapixel Sony IMX500 sensor that can capture images at resolutions of 4056 x 3040 pixels at 10 frames per second or 2028 x 1520 pixels at 30 frames per second. With a 78-degree field of view and manually adjustable focus, it's suitable for a variety of projects.
The sensor's integrated AI accelerator allows the camera to execute machine learning models efficiently, making it possible to perform tasks like object detection and pose estimation directly on the camera.
An onboard RP2040 microcontroller manages neural network operations and firmware updates, providing support for models developed using frameworks like TensorFlow and PyTorch. This setup enables developers to deploy custom-trained models tailored to their specific needs.
Compatibility and performance across Raspberry Pi models
One of the key advantages of the new AI Camera is its compatibility with all Raspberry Pi models, including the tiny Raspberry Pi Zero. It connects via standard ribbon cables, making it an accessible upgrade for existing projects. The camera's low power consumption and reduced latency are particularly beneficial for applications that require continuous operation, such as surveillance systems or autonomous robots.
By handling AI processing on the camera itself, the Raspberry Pi's main processor remains free for other tasks, improving overall system performance. This is especially useful for older or less powerful Raspberry Pi boards that may struggle with heavy computational loads.
How it compares to the Raspberry Pi AI Kit
Earlier this year, Raspberry Pi released the AI Kit, a powerful accessory capable of performing thirteen trillion operations per second. However, the AI Kit is only compatible with the Raspberry Pi 5 and requires a separate camera module, adding to the total cost and complexity. The new AI Camera offers a more streamlined solution by combining image capture and AI processing in a single device that's compatible with all Raspberry Pi models.
While the AI Kit delivers higher theoretical performance and supports a broader range of models, the AI Camera provides sufficient AI acceleration for most typical projects at a lower cost. Developers can choose between the two based on their specific requirements, such as the need for higher performance or broader model compatibility.
The camera integrates seamlessly with the Raspberry Pi camera software stack, allowing developers to use tools like libcamera, rpicam-apps, and Picamera2 to process images and neural network outputs. This tight integration simplifies the development process and enables efficient handling of tasks like real-time object detection and pose estimation.
