Assessment of Efficient and Cost-Effective Vehicle Detection in Foggy Weather

Efficient and cost-effective vehicle detection under foggy and smoggy weather conditions remains a significant challenge in the domains of Intelligent Transportation Systems (ITS), Advanced Driver Assistance Systems (ADAS), Collision Avoidance Systems (CAS), and traffic surveillance. Vehicle detection in foggy weather, a critical requirement for preventing road collisions and ensuring driver assistance, has not been adequately evaluated. The state-of-the-art deep learning algorithm, YOLO-V11 network architecture, is studied and evaluated for efficient and cost-effective vehicle detection in diverse foggy weather road driving conditions. To our knowledge, YOLO-V11 performance has not been evaluated in diverse foggy weather road driving conditions. The vehicle Detection in Adverse Weather Nature (DAWN) and Foggy Driving (FD) datasets are utilized for the performance evaluation across diverse object detection performance metrics. The DAWN and FD datasets are customized and self-annotated to four vehicle detection classes: buses, cars, motorcycles, and trucks. Extensive experiments are performed to assess the efficiency and cost-effectiveness of YOLO-V11 models on DAWN and FD datasets. YOLO-V11s model outperformed with 73.1 % mAP and 71 % F1 score on the DAWN dataset and YOLO-V11m outperformed with 47.7 % mAP and 47 % F1 score. The recall and precision score of YOLO-V11m is higher on the DAWN dataset. Whereas, the recall score of the YOLO-V111 model is higher on the DAWN dataset and the precision score of YOLO-V11n is higher on the FD dataset. YOLO-V11 outperformed with an inference speed of 26 FPS on the DAWN dataset and YOLO-V11n outperformed with an inference speed of 13 FPS on the FD dataset. The comparative analysis concludes that the YOLO-V11n and YOLO-V11s models detect vehicles with faster inference speeds while maintaining high accuracy. Conversely, YOLO-V11m and YOLO-V111 models detect vehicles with enhanced accuracy but at the cost of slower inference speeds.