Raspberry Pi Teams Up With Sony to Launch AI Camera

The new AI camera integrates an intelligent vision processor and can be used with Raspberry Pi single-board computers.

Sony Semiconductor Solutions (SSS) and Raspberry Pi have announced a new AI camera with a built-in AI processor. The camera, compatible with all Raspberry Pi computers, is built around Sony’s IMX500 intelligent vision sensor.

Raspberry Pi AI Camera jointly developed with Sony Semiconductor Solutions. 

The IMX500 has built-in AI processing to offload AI functionality from the Raspberry Pi. Sony has collaborated on edge AI with Raspberry Pi since taking a minority stake in the single-board computer (SBC) developer in April 2023. The two companies have been working since then to develop an edge AI platform for Raspberry Pi developers.

AI in the Maker Space

The Raspberry Pi was first released in 2012 as a low-cost educational SBC. Since that time, the organization has released more than a dozen models and a self-designed microcontroller. The SBCs run on Linux, most other operating systems, and Arm-compatible processors.

The Raspberry Pi Foundation's mission is to help educate students and others about computing and how to create with digital computers. With this goal as its driving force, Raspberry Pi computers are intuitive enough for users with limited computer experience. Still, they are powerful and well-documented enough for advanced enthusiasts and professionals to use as well.

By putting the IMX500 AI Camera into the hands of the Raspberry Pi organization and user community, AI image processing will be available to far more people than it would be if offered on just about any other platform.

Key Features of the Raspberry Pi AI Camera

The camera (datasheet linked) uses an IMX500 intelligent vision sensor with on-chip edge processing. The intelligent sensor is designed to capture images and conduct AI processing before sending data to the host computer. The sensor is built in a stacked configuration with an on-chip SRAM and a digital signal processing (DSP) engine.

IMX500 intelligent image sensor.

Other key features of the AI camera include: 

• 4056 x 3040 resolution (approximately 12.3 megapixels)
• 7.857-mm diagonal with a 1/2.3 aspect ratio
• 1080p video at up to 240 fps
• Full video with AI processing at 30 fps
• On-chip AI processing, ISP, and HDR video
• Tensor input size from 64 x 48 to 640 x 480 (int8 or uint8)
• 8.4 Mbytes of memory for firmware, network weight file, and working memory
• Full camera dimensions: 25 mm × 24 mm × 11.9 mm with mounting holes compatible with the Raspberry Pi Camera Module 3

The camera captures both video and still images with tensor metadata without needing help from the Pi host processor. This functionality enables powerful AI vision systems without specialized coprocessors. Software support for tensor metadata is already available in libcamera and Picamera2 libraries and the rpicam-apps application suite.

Raspberry Pi Support

Support has always been one of the strong suits of Raspberry Pi, and this camera is no exception. As with other Pi hardware, the AI camera comes with extensive setup and application documentation.

Output of the object-detect example application.

Installation will be familiar to any Raspberry Pi user, with just a single Linux command line operation to complete the installation. Post-setup operation can begin right away with the use of included examples.

AITRIOS Development Tools

The Pi AI Camera is not just for maker and hobby systems. It is compatible with Sony’s AITRIOS AI sensing development platform. AITRIOS is a commercial edge AI platform for the Internet of Things (IoT) vision development. AITRIOS is designed to securely handle large volumes of data with reduced latency in data transmission, processing, and analysis. AITRIOS combines support for various input devices with application-specific services and cloud support.

How to Get a Raspberry Pi AI Camera

The camera is available now at authorized Raspberry Pi resellers and is planned to remain in production until January 2028. Software is available from the Raspberry Pi website, the Pi GitHub site, and a number of third-party contributor GitHub sites.