Low Power Sensor Fusion Targeted for AI Applications at The Edge

Threat detection and physiological monitoring of soldiers from fused sensor data collected in real time is currently limited to running deep neural networks with substantial computing needs. The lack of data acquisition from sensor readings and efficient detection of novel enemy signatures motivates the need for a low-power, low-cost, wireless multi-sensor fusion computing system. We propose the current trends in Internet of Things to deploy a chargeable, wireless multi-channel acquisition system that can be interfaced with a high speed, Single Board Computer (SBC) such as the NVIDIA Jetson Orin capable of running object detection models, such as YOLOv7-tiny to enable high speed target detection, and health monitoring, at a low-cost, and low-power. Target detection and data fusion was achieved at 60 FPS with a YOLOv7-tiny model trained on a custom drone dataset with a NVIDIA Jetson Orin equipped with a USB camera, a MSP430FR2355 interfaced over UART with fused data from two I2C sensors, and two ADC sensors. Based on the power metrics measured with the MSP430 and the interfaced sensors, a multi-channel acquisition system was designed that features a micro-USB battery charging interface capable of charging aLi-Ion battery (400 mAh) to power the system.