This document proposes an object detection technique for aerial videos based on motion vector compensation and statistical analysis. It begins with an introduction to the importance of object detection in aerial surveillance. It then describes the characteristics of aerial video images and a preprocessing method using Bayesian wavelet denoising. A compensation of motion vectors is performed using camera motion estimation. Statistical analysis and clustering of compensated motion vectors is used to detect objects and eliminate isolated vectors. The method is tested on a road surveillance video, showing it can effectively detect objects after noise removal and motion vector processing.