Enhanced Vehicle Movement Counting at Intersections via a Self-Learning Fisheye Camera System

Abstract

Accurate vehicle counting at intersections is crucial for assessing traffic flow and gaining insights into vehicle trajectories captured by traffic cameras. This paper introduces an innovative framework that leverages a fisheye camera system to count vehicle movements at intersections with two significant contributions: First, the proposed algorithm employs a novel zone-based counting methodology to categorize and collect trajectory data and autonomously learn movement patterns with the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm at intersections. Second, as the algorithm becomes proficient in recognizing the paths traversed by vehicles, it seamlessly transitions into a hybrid mode, integrating both zone-based and path-based counting techniques. It enables accurate vehicle counting even in challenging scenarios involving broken tracks or partial trajectories. The self-learning capability of the proposed method enhances its flexibility and scalability, enabling it to adapt to diverse traffic patterns at different intersections without manual intervention. The performance of the proposed method is accessed by conducting experiments on three real-world fisheye camera footage datasets. The counting results underscore the efficacy of our innovative approach, showcasing an outstanding F1 score surpassing 98% across all evaluated intersections. This performance highlights its potential for real-world applications, including intelligent traffic signal control, urban planning, and emission estimations within traffic management frameworks.

Keywords

• Object detection
• trajectory clustering
• traffic monitoring
• trajectory similarity
• vehicle counting