Weather Resilient Vehicle Detection and Tracking System based on YOLOv7 and Faster-RCNN Algorithm

Abstract

Further development in vehicle detection algorithms using deep learning necessitates reliability on the opposite end of broad behavioural conditions affecting their performance. This study assesses four such state-of-the-art algorithms -Faster Regression-based Convolutional Neural Networks (R-CNN), Histogram of Oriented Gradients (HoG) and You Only Look Once (YOLO) not solely on their construction but on performance in vehicle detection on these technologies. The work is supplemented with some statistical validation techniques, such as confidence intervals and analysis of variance (ANOVA). It was statistics-based work that proved YOLO to be better than Faster R-CNN, showing mean average precision of 78.9 for foggy and rainy conditions including an inference time of 12 ms/frame. Furthermore, precision-recall analysis has demonstrated YOLO's ruggedness to cover small and obstructed vehicles.