Human Detection by Omni-directional Camera Semester Thesis

We present a human detection system based on an omnidirectional camera which has the advantages of detecting a large view-field. We adopted Haar-like feature based AdaBoost cascade classifiers to detect different human body parts. A proper probabilistic model for presenting the arrangement of the detected human body parts have been proposed in this paer, and successfully reduced the false positive alarms. We combine the tracking algorithms with the detection method into a same framework. The performance of recognition has been greatly improved by using this method.