Real-time deep learning pedestrians classification on a micro-controller

Abstract

Deep learning neural network is one of the most advanced tools for object classification. However, it is computationally expensive and has performance issues in real time applications. This research’s use-case is efficient design and deployment of deep learning neural networks on palm sized computers like Raspberry Pi (RPi) as an in-vehicle-monitoring-system (IVMS) for real-time pedestrian classification. I have developed a system based on a neural network template named Cafenet that runs on an RPi and can classify pedestrians using deep learning. Simultaneously, I have proposed a new classification system based on multiple RPi boards, which offers users two modes of pedestrian detection: one is fast classification, and the other is accurate classification. The experiments results show that the device could classify pedestrians in real-time and the detecting accuracy is acceptable.