Detection and Recognition of Vehicles Using Indian Driving Datasets

• Published in: 2023 IEEE International Conference on Cloud Computing in Emerging Markets (CCEM) pp. 105 - 109
• Main Authors: S, Yathish Padukote H., R, Suresha, N, Manohar, Jipeng, Tian
• Format: Conference Proceeding
• Language: English
• Published: IEEE 02.11.2023
• Subjects: Accuracy; Computer vision; Deep learning; Faster RCNN; Image recognition; Indian Driving Dataset (IDD); Lighting; Powel Law Transformation; Road traffic; Surveillance; Testing; Traffic control; Vehicle detection; Vehicle recognition
• ISSN: 2573-2986
• DOI: 10.1109/CCEM60455.2023.00024

Vehicle identification and recognition are essential computer vision tasks with important applications in autonomous driving, traffic management, and surveillance systems. The Indian Driving Dataset (IDD) dataset used in this work was compiled from a variety of driving scenarios in the Indian environment, and it includes a thorough investigation on vehicle detection and recognition. The dataset is meticulously chosen to reflect the difficulties brought on by the state of Indian roads, traffic patterns, and vehicle kinds. In this study, a Faster RCNN method for precise and effective vehicle detection and recognition deep learning approaches by enhancing images using power-law transformation. The collection consists of annotated IDD dataset images taken in a range of lighting situations, urban-rural locales, and camera angles. The testing results demonstrate the effectiveness of the recommended approach, which achieves competitive performance with 79.5% detection accuracy and 76% recognition accuracy. Additionally, the performance of the model is impacted by elements like various lighting situations and various vehicle models. This research offers insightful information about the model's ability to generalise in realistic situations.