For real-time AI applications, it is important to process incoming data using a trained neural network model. For faster data processing, the introduction to AI inference at cloud did support many aspects, increasing efficiency, but the need to process the data at the end device has become vital. To serve this purpose, we have AI inference at the edge. With AI inference at the edge, the market for real-time data processing and obtaining quick results has increased performance. When it comes to AI inference at the edge, several ASICs in the market are custom-built for this purpose. Google introduced an Edge TPU application-specific integrated circuit designed to run inference at the edge.
There have been several designs using Google’s Edge TPU, but this specific project, Maple Syrup Pi Camera, an AIoT smart camera has caught the eye of the makers’ community due to its use of Raspberry Pi Zero W for AI inference at the edge. We don’t usually see these SBCs used to run inference at the edge due to their low processing capabilities. For specific AI purposes, we have several powerful SBCs in the market designed for this use case. [Ricardo] has developed this smart camera that uses a Coral USB accelerator that is connected with the Raspberry Pi Zero W.
This interesting project of exploring the power of Raspberry Pi Zero W has employed the Raspberry Pi Camera along with the Raspberry Pi Zero flat cable. The software provided by the designer on GitHub, recommends you to use 8GB of storage Micro SDCard but it depends on the firmware you will be using. Since the SBC does not come with a power protection circuit, the board connects the USB power directly to the power supply. “That means the Coral USB Accelerator will be directly connected to the power supply, allowing it to drain as much current as of the power supply and impedance of microUSB + PCB traces allow it,” [Ricardo] explains:
In my experiments, the Maple-Syrup-Pi-Camera consumes around 160mA at 5V when idle (800mW).
This exploration of machine learning capabilities could be a good start for those looking to use Google Coral USB Accelerator with their favorite Raspberry Pi. Such projects can be used for object detection or even processing the incoming data on edge and giving the required output through trained models. Such projects can be used in the area of real-time AI-vision applications. For more information on this open-source project, head to the GitHub repository.