Crowd Counting

Crowd counting studies how to estimate the number and spatial distribution of people in images and video, which matters for monitoring transport hubs, public events, and other crowded spaces. Common approaches predict density maps or points, detect or segment individual people, and combine visual features with depth, thermal, or other sensor cues; recent work also explores foundation models and synthetic training data. Methods are commonly evaluated on datasets such as ShanghaiTech, UCF-QNRF, and NWPU-Crowd, as well as in aerial and multi-camera settings. A central challenge is making counts reliable across changes in camera view, lighting, and crowd density, especially in extremely dense scenes where standard benchmarks and annotation methods may not reflect deployment conditions.