On-line Learning a Person Model from Video Data
Peter M. Roth, Horst Bischof · 2006
Recently we have proposed the Conservative Learning Framework [2] for unsupervised learning a person detector from video data. The main idea is to minimize the manual effort when learning a classifier and to combine the power of a discriminative classifier with the robustness of a generative model. Starting with motion detection an initial set of positive examples is obtained by analyzing the geometry (aspect ratio) of the motion blobs. If a blob fulfills the restrictions the corresponding patch is selected. Negative examples are obtained from images where no motion was detected. Using these data sets a first discriminative classifier is trained using an on-line version of AdaBoost [1]. In fact, applying this classifier all persons are detected (we got a quite general model) but there is a great number of false positives. Thus, we apply a generative classifier, i.e., robust PCA [3] to verify the obtained detections