Exploring different dehazing algorithms for object detection in foggy weather conditions for autonomous vehicles

• Main Author: Kim, Chae Yoon
• Other Authors: Lu Shijian
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Object detection
• Online Access: https://hdl.handle.net/10356/174978
• Institution: Nanyang Technological University

This report presents a comprehensive analysis of the performance of various dehazing algorithms in the context of object detection for autonomous vehicles operating under foggy weather conditions. With the advent of autonomous driving technology, the need for reliable object system in diverse conditions, including adverse weather, has become paramount. Fog significantly degrades the visibility of road scenes, challenging the object detection modules of autonomous vehicles by introducing noise and reducing the contrast and clarity of input images. In this report, several state-of-the-art dehazing algorithms — Dark Channel Prior, Multi Scale Optimal Fusion, and Color Cast Dependent Dehazing — are evaluated for their effectiveness in enhancing the accuracy of object detection systems, specifically Faster R-CNN, under foggy conditions. These dehazing methods are integrated with Faster R-CNN and each model’s performance is assessed using various object detection metrics, indicating each method’s capability in enhancing object detection in foggy images. The findings of this report underscore the critical role of dehazing as a preprocessing step in the object detection pipeline of autonomous vehicles, particularly in foggy weather conditions. By providing insights into the strengths and limitations of different dehazing algorithms, this study aims to guide the development of more resilient object detection systems that ensure the safety and reliability of autonomous driving in adverse weather conditions.