Method and procedures for automated counting of number of persons (or objects) in a scene monitored by a surveillance camera. A semi-supervised procedure (active learning) for constructing a count distribution manifold to extrapolate labels (inference of counts) from very large unlabelled data points with the assistance of human-labelled counts from a very small labelled data set. Receiving a label, signal or other indication from a user. Identifying the most informative data point or points with a known label (count). Constructing a regression model using the user labelled data and inference-labelled data. With or without a user's selection of targeted zones in a scene and targeted unseen video data, inferring/predicting (estimating) the counts of persons (or objects) in the unseen data in the entire scene or in the selected zones individually and/or accumulatively.
展开▼