Wide-angle scenes
Modeling dense, all-direction pedestrian layouts across a large field of view and extreme edge distortion.
Overhead perception research
See the scene
from a wider angle.
Researching robust pedestrian detection for overhead fisheye imagery—from geometric distortion and oriented targets to efficient, real-time perception.
FOV · 180°LIVE · LAB 0426
FISHEYE PERCEPTIONDISTORTION-AWARE VISION
180°Field of view
360°Scene awareness
OBBRotated detection
RTReal-time goal
01 · RESEARCH FOCUS
Built for difficult views.
Overhead fisheye scenes compress scale, bend geometry, and rotate every subject. Our work treats those effects as useful priors.
Oriented detection
Using rotation-aware localization to follow the natural radial orientation of overhead pedestrians.
Efficient inference
Balancing accuracy and computation for lightweight, practical, and real-time deployment.
吃→
CAMPUS TOOL · 01
Research is hard. Lunch shouldn’t be.
大美、至美、小美、天美——让随机选择器替你结束饭点纠结。支持食堂筛选、楼层偏好、二选一 PK 和饭后小挑战。
02 · PIPELINE
From pixels to perception.
01 · CaptureOverhead wide-angle imagery
02 · CalibrateGeometric and radial priors
03 · DetectMulti-scale oriented targets
04 · DeployEfficient real-time inference