Video Analysis of Vehicle Detection and Shadow Removal with Gaussian Mixture Model

Abstract

Vehicle detection plays a vital role in Intelligent Transportation Systems (ITS). However, shadows present a significant challenge to the accuracy of traffic monitoring systems, often leading to the misclassification of shadows as vehicle components, distortion of object shapes, and compromised detection precision. Addressing this issue is crucial for enhancing the performance of ITS, as effective shadow detection and removal can improve vehicle detection accuracy, optimize traffic flow management, strengthen safety measures, and provide more reliable data for informed decision-making. This study proposes an improved approach to shadow removal that integrates the Gaussian Mixture Model (GMM) for vehicle detection with shadow removal using the HSV colour model, further refined by the Graph Cuts algorithm for improved segmentation. The video dataset used in this study is from the public Kaggle repository. The initial stage of shadow removal utilized the HSV colour model to process foreground features, followed by frame segmentation with Graph Cuts to eliminate residual shadow outlines, addressing the limitations of the colour model-based method. Comparative analysis revealed that incorporating Graph Cuts significantly enhanced shadow removal performance. The proposed algorithm achieved an average shadow detection rate of 92.10% without Graph Cuts and 98.25% with Graph Cuts. Furthermore, the method consistently maintained shadow discrimination rates exceeding 99% across all vehicle colours. These experimental results underscore the efficacy of the proposed framework in eliminating vehicle shadow outlines and enhancing the accuracy of vehicle detection, offering a robust solution for ITS applications.