Embedded and real time vehicle classification system with occlusion handling.

The idea of applying AI in embedded systems is growing in all sectors, from airplanes to drones, cars, cell phones and robots. The challenge for this type of embedded systems is to ensure real-time functioning, with accurate results under a robust system. In this paper, we propose an embedded and real-time road traffic classification system based on AI techniques. The software part includes the different blocks dedicated to vehicle counting and classification, which is performed by using background subtraction, Histograms of Oriented Gradients and Support Vector Machine. The proposed method also includes a set of linear and non-linear filtering techniques for noise removal. The occlusion handling is applied during classification process by training the model on classes containing occlusion images. The hardware part of the system consists of Raspberry Pi board with an operating system installed for flexibility purposes. The implemented algorithm has been improved to guarantee a real time operation and a minimal storage in the memory. The proposed system achieved a counting accuracy of 96.34%, and a processing speed around 4.55 FPS, which is adequate with all the performance and real time criteria. [ABSTRACT FROM AUTHOR]