Enhancing Traffic Counting in Rainy Conditions: A Deep Learning Super Sampling and Multi-ROI Pixel Area Approach

Abstract

In Intelligent Transportation Systems (ITS), adaptive traffic control relies heavily on precise, real-time traffic data. Controllers use information such as vehicle count, vehicle density, traffic congestion, and intersection wait times to optimize traffic flow and improve efficiency. Traffic cameras collect and process this data, but environmental factors like rain can degrade the performance of data retrieval systems. We propose a vehicle detection method that integrates pixel area analysis with Deep Learning Super Sampling (DLSS) to enhance performance under rainy conditions. Our method achieved an accuracy of 80.95% under rainy conditions, outperforming traditional methods, and performing comparably to specialized methods such as DCGAN (93.57%) and DarkNet53 (87.54%). However, under extreme conditions such as thunderstorms, the method's accuracy dropped to 36.58%, highlighting the need for further improvements. These results, evaluated using the AAU RainSnow Traffic Surveillance Dataset, demonstrate that our method improves traffic data collection in diverse and challenging weather conditions while identifying areas for future research.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint