Detection-based video surveillance using Deep Neural Networks on STM32 microcontroller
Alessio Canepa, Edoardo Ragusa, Rodolfo Zunino, Paolo Gastaldo · 2022 29th IEEE International Conference on Electronics, Circuits and Systems (ICECS) · 2022
Object detection is one of the most active research areas in the computer vision field. Using object detection techniques, nowadays mostly based on deep neural networks, new intelligent camera-based surveillance systems can be designed, capable of generating alerts only in the presence of specific objects, like persons, in the camera field of view. However the memory and computational load required by these techniques makes it challenging to use them on low power, miniaturised and resource constrained surveillance devices designed for harsh environments. In this paper, we show an efficient method to detect the presence of a specific object in surveillance video frames using deep neural networks on an STM32 microcontroller, suitable for harsh environments. Our solution achieved 97% precision and 93% recall, while consuming less than 400 mW.