For designers, the demand for IoT devices often translates to a demand for ultra-low-power embedded designs. This stems from the fact that IoT devices tend to be battery-powered, but often integrate a plethora of different technologies onboard—from sensors to RF to AI processing.
The RSL10 “Smart Shot” camera. Image used courtesy of ON Semiconductor
One embedded system that can be particularly challenging is the market-termed “smart camera.” These systems require solid-state image sensors, optics, and computer architecture while implementing computer vision technology for machine vision. Supporting all of these functions in a battery-powered device while maintaining a small form-factor introduces many engineering tradeoffs.
To address these tradeoffs, ON Semiconductor recently released a new event-triggered smart camera.
One of the main tricks to minimize power consumption in IoT devices is to idle the device as much as possible; the device turns off automatically when it isn’t being used to conserve battery life. This concept equates to an event-triggered camera—cameras that only capture images when provoked by some event.
These devices often employ AI inference along with information from the environment to help determine when an event should trigger the camera. The specific trigger will change from application to application, but common triggers could include timers, motion sensors, accelerometers, and temperature.
Basic structure of a smart camera, including an ASIP (application-specific information processor). Image used courtesy of Fábio Dias Real et. al and ResearchGate
From a hardware perspective, these types of systems normally require a camera, RF ICs for connectivity, sensors for data acquisition and for camera-triggering, and some sort of processing unit like an MCU.
RSL10 at the Board Level
At the board level, ON Semiconductor’s RSL10 leverages a few power and size advantages that may be useful to embedded designers.
For the camera, the board uses an ARX3A0 Mono 65-degree DFOV IAS Module. This CMOS-based image sensor is designed to be very low power, drawing less than 3.2 mW with a 60 fps rate while active. ON Semiconductor also says the camera has built-in capabilities to detect motion or changes in the environment’s lighting—a boon for event-triggered recording.
Component-wise breakdown of the RSL10 camera PCB. Image used courtesy of ON Semiconductor
For RF connectivity, the device employs an RSL10 system in package (SiP). According to the datasheet, the SiP contains an “onboard antenna, RSL10 radio SoC, and all necessary passive components in one package to help minimize overall system size.” In the context of this design, the SiP is used for BLE connectivity (Bluetooth 5) to connect to the cloud, offloading data for AI processing.
On the user interface (UI) side, the camera comes with a companion app where users can remotely configure sensors, modify image triggers, and view data from the cloud. Connectivity to the app is also done via BLE.
Battery Life vs. Edge Processing
ON Semiconductor claims that this new camera is so low power that its battery life can last longer than a year. This does, however, come at the cost of offloading AI performance to the cloud as opposed to edge processing.
According to the company, applications for the camera may include wearable safety equipment, automated checkout support at stores, occupant monitoring in vehicles, and more.