Moving Vehicle Logo Detection based on YOLOv7 under Complex Lighting Conditions

Vehicle logo detection is an important component of the Intelligent TransportationSystem (ITS) that has received a lot of attention from researchers.Its ability to provide supplementary information enhances the process of vehicle identifications, making it an essential tool in a variety of ITS applications such as parking management, traffic monitoring, toll collection and autonomous vehicles and advanced driver assistance systems (ADAS).These applications require the vehicle logo detection to work not only in an optimal lighting environment but also in complex lighting conditions as well.Our work focuses on a deep learning-based vehicle logo detection model that can detect logos under various lighting conditions such as low lighting, strong lighting, blurriness, and glares.A dataset of surveillance footage consisting of 10 common car brands in Malaysia is created for model training and testing.Furthermore, this study involves a comparative performance analysis of two deep learning models, namely YOLOv3 and YOLOv7 with image enhancement method, to evaluate their suitability for logo detection.Our findings reveal that YOLOv7 exhibits a performance improvement of 0.228% compared to YOLOv3 when applied to the complex lighting conditions dataset with the application of contrast limited adaptive histogram equalisation (CLAHE) as an image enhancement technique.