System and method for automatic person counting and detection of specific events
View Patent ↗A system for detecting and counting individuals in a stationary or moving crowd based on a digital or digitized image captured from a single camera. Initial information is assumed based on a foot-to-head plane homology, where a geometric construct is developed to best enclose image features with a high probability of being an individual within a crowd. These geometric constructs are then subjected to further probabilistic analysis to determine individuals. The vector track of each individual is determined to validate the determination of a group of features as individuals, thereby compensating for occlusion of an individual within any given frame image. A virtual gate is then employed to count the individuals moving past the virtual gate. Also, by weighting portions of the individual, events near the gate can be predicted.
1. A method for counting at least one individual in a crowd comprising the steps of:
(i) calibrating an image capturing device and an image processing device;
(ii) capturing a flame of an image with the image capturing device;
the image processing device performing the steps comprising:
(iii) drawing a virtual gate through the image;
(iv) applying a foreground estimation module to the image;
(v) applying a tracking module to the image;
(vi) applying a crowd segmentation module to the image to separate the individual from the crowd;
(vii) counting the individual;
wherein the crowd segmentation module in step (v) comprises the steps of
(i) identifying at least one low level image feature;
(ii) hypothesizing at least one clique from the low level image features; and
(iii) constraining the clique according to a geometric shape model;
wherein the constraining of step (iii) comprises:
(iv) assigning each low level image feature to a single clique;
(v) establishing future assignments according to an assignment vector based on a likelihood function;
wherein the likelihood function is a free energy equation.
2. The method of claim 1 , wherein the calibration of step (i) utilizes foot-to-head plane homology.
3. The method of claim 1 , wherein the tracking module comprises applying the Hungarian algorithm.
4. The method of claim 1 , wherein the tracking module and the segmentation module are integrated.
5. The method of claim 1 , wherein the individual comprises a person.