Automotive Embedded Image Classification Systems

Abstract

Electric vehicles are becoming more autonomous, so they must classify images using embedded systems and advanced classification methodologies to achieve a fast response when navigating. Thus, studying and analyzing classification algorithms and embedded systems is a mandatory endeavor to improve the performance of electric vehicles during their operation. On the other hand, artificial intelligence is one of the leading technology topics in autonomous electric vehicles; however, the computational requirements to analyze a large amount of data in real-time would mean having costly and powerful computers on board. Also, this can mean using a significant physical space in the vehicle and energy resources. An embedded system can handle the necessary data to classify standard traffic signals on the road so the principal processor can be released from these tasks. This paper proposes a traffic signal object detector and classifier that is implemented using a Tiny YOLOv4 and compares Frames Per Second obtained in an embedded system using the trained model, a web camera, and a Hardware Accelerator called Movidius Neural Stick by Intel are integrated into the proposed solution. The results show that the proposal is a good alternative for implementing a specialized image classification system into an embedded digital system for electric vehicles. This proposal could be extended to classify more images that can show up on a conventional road.