Efficient Vehicle Detection and Classification Algorithm Using Faster R-CNN Models

Abstract

This work focusses the issue of traffic analysis of video surveillance using Efficient-RCNN. The challenge here is expressed as classifying and counting vehicles in Traffic. To provide this proposed issue, we used the related work of efficient RCNN and detector vehicles. For the classification in live traffic movement. The robustness of the efficient RCNN was provided by many restrictions such as adaptive feature pooling, anchors optimization, focal loss, and other mask branch. For training the proposed model, we used UA-DETRAC dataset of ten hours of videos at twenty-five frames / seconds, with resolution of ninety hundred sixteen× five hundred fourteen pixels, this dataset has more than one hundred fourteen k frames in the UA-DETRAC dataset and eight Townend two hundred fifteen vehicles that are manually annotated with DarkLabel, we have 80% recall for proposed method compared to 79% for YOLO, and we also have 79% of precision for the proposed method compared to 78 in YOLO