Automatic Crowd Detection Based on Unmanned Aerial Vehicle Thermal Imagery

Automatic crowd detection is one of the most challenging problems in computer vision. Although the thermal image from Unmanned aerial vehicle (UAV) have prominent performance in perspective, but have some disadvantage, such as the platform motion, image instability and so on. In the thesis, the automatic crowd detection system is designed to tackle these disadvantage. The detection system consists of two parts: ROI extraction and SVM classification, circle gradient and geometric filtering are used in ROI extraction, a hybrid feature which combines circle gradient (CG) and histogram of oriented gradient (HOG) is used in SVM classifier. The experimental results prove that the approach performs dense crowd detection from thermal images effectively.