Improving Traffic Object Detection with Vehicle-Borne Fisheye Imaging

The primary objective of this project is to develop a robust and accurate traffic object detection framework for fisheye camera images by designing and implementing YOLO26-SPAC and YOLO11-CFA architectures that incorporate attention mechanisms capturing spatial, frequency, and multi-scale contextual features. The models are trained on the FE8K dataset to classify five traffic object categories: Bike, Bus, Car, Pedestrian, and Truck, with feature enhancements including spatial pyramid attention, cross-frequency attention, wavelet feature extraction, and global context embedding to improve detection accuracy and robustness against geometric distortions. The system’s performance is evaluated using metrics such as precision, recall, F1-score, and mAP, supported by visualization through confusion matrices, PR curves, and validation prediction grids. A modular framework integrating front-end and back-end components facilitates user interaction, image input, and prediction output. The framework is designed to handle variations in image resolution, object scale, and camera distortion efficiently, providing a scalable and reliable solution for accurate traffic object detection in fisheye imagery.