Multi-Frame Object Detection.

摘要

This thesis describes an object detection system that extracts and combines appearance information over multiple consecutive video frames, inherently gaining and analyzing information related to motion. Objects that exhibit characteristic motion over the course of multiple frames are able to be detected at smaller scales than is achievable using a single-frame detector. Our method builds on the detection work of Viola and Jones, with our extension being the added ability to combine information from multiple frames. Our implementation detects an object in synthetic images at very small scales, down to 3x3 pixels, and has a low false-alarm rate.