Intersection Analysis Using Computer Vision Techniques with SUMO

Abstract

This paper presents intersection analysis using computer vision techniques with Simulation of Urban MObility (SUMO). At first, an efficient deep-visual tracking pipeline is proposed by using the off-the-shelf YOLO object detection architecture and cascading it with a discriminative correlation filter (CSRT) to produce reliable trajectories for traffic analysis of vehicles and pedestrians. While a variety of traffic measurements can be directly estimated from the extracted trajectories (e.g., speed, turning movement count), a method of incorporating turning movement count (TMC) within SUMO is proposed in order to mimic a realistic traffic flow for an observed intersection and its comprehensive analysis. Experimental evaluations on developed tracking system implies that YOLOv5 variant is the best for traffic cameras and after appropriate fine-tuning using the UNLV Pedestrian data-set, the YOLOv5 performance manifested a significant improvement with value of 0.62 in recall value. The tracking system is further employed for monitoring three other intersections in the downtown of Las Vegas and turning movement counts were estimated for peak hours of morning and evening time of one day 7:00-9:00 and 16:00-18:00) with 15 minutes intervals. Finally, the intersection design including traffic signals with estimated TMC are used to calibrate SUMO to provide critical parameters (e.g., lane density, travel time, occupancy) for traffic signal performance evaluation and comprehensive intersection analysis. The signal design treatment demonstrates significant improvement for travel times and simulation results indicates that turning left ratio is a crucial factor affecting the travel time of vehicles on each intersection leg.