Robust Vehicle Detection Using Image Processing

Abstract

ABSTRACT The escalation of vehicular traffic in a country like Sri Lanka has escalated concerns over traffic congestion and safety. This has prompted a critical need for robust vehicle detection systems to facilitating the implementation of effective traffic management strategies and understand traffic patterns. This research addresses the challenge of vehicle detection and identify vehicular categories in Sri Lankan contexts, where reliance on private transportation is leading due to the absence of an advanced public transport infrastructure. Eventhough the existence of methodologies like the Viola-Jones object detector, extracting moving vehicular feature array for vehicle detection remains a formidable task in traffic analysis. To address this problem, we propose a novel approach leveraging image processing algorithms, particularly focusing on image segmentation. Novel algorithm demonstrates a remarkable ability to extract irregularly shaped objects from traffic video streams, achieving a highest success rate in segmentation accuracy. By using the video data collected at Kottawa highway bridge and Kotte- Thalawathugoda Road using digital cameras, our methodology captures real-time traffic data and categorizes vehicles into distinct classes such as motorcycles, three-wheelers, cars, vans, lorries, and buses. Furthermore, we emphasize the importance of accurate traffic information in optimizing mobility at urban intersections, which is crucial for all road users. The proposed system not only offers insights into traffic dynamics but also contributes to the development of smart and sustainable mobility systems. However, challenges such as environmental noise, sudden illumination changes, and shadow effects necessitate further research to enhance the robustness and reliability of the proposed methodology. In conclusion, this thesis presents a comprehensive approach to address the complexities of vehicular traffic analysis in Sri Lanka, offering valuable insights for researchers, stakeholders, and policymakers. The proposed method gives a success rate of more than 95% in irregular shaped image segmentation. Results of vehicle extraction from the video sequence gives average accuracy of 94.1%. In this research six vehicle types have been considered for the results.