A research project for overhead fisheye image analysis, rotated pedestrian detection, geometric distortion modeling, and lightweight visual perception.
Designed for overhead fisheye images with large field of view, strong distortion, and dense pedestrian distribution.
Uses oriented detection to better locate pedestrians under fisheye deformation and complex viewing angles.
Focuses on efficient inference, fewer parameters, lower computation, and real-time deployment potential.
This project explores scale reallocation, fisheye geometric priors, feature fusion, and efficient detection methods for robust pedestrian detection in overhead fisheye images. The goal is to improve the detection accuracy of small, edge-distorted, and densely distributed targets while maintaining real-time performance.