Counting, Classifying and Tracking Vehicles Routes at Road Intersections with YOLOv4 and DeepSORT

Abstract

The decision to install traffic signs at intersections, safety lanes or even widening areas is always subject to a technical survey. The collection of local images with cameras and the subsequent analysis is a useful procedure in these situations. This article presents a method for analyzing video images of road intersections for the detection and tracking of four categories of vehicles: cars, buses, trucks and motorcycles. The preliminary results of vehicle route accounting show that the methodology is quite promising, achieving good results in several cases, but much research is still needed to make these systems robust to occlusions and other unexpected events.