CROWD BEHAVIOR RECOGNITION METHOD AND DEVICE BASED ON STRUCTURE LEARNING

摘要

The present invention relates to a crowd behavior recognition method using a computer. The present invention is to provide a crowd behavior recognition method based on structure learning, capable of improving accuracy and usefulness by automatized crowd behavior learning, and a device for the same. The crowd behavior recognition method based on the structure learning using the computer includes: (i) a step of extracting an image feature vector composing multiple elements by unit image, from an image signal including the unit image obtained every image sections; (ii) a step of determining a crowd behavior class about the image feature vector in the image section by inputting the image feature vector extracted during the image section in a classification device which is set to determine one crowd behavior class based on multiple elements of the image feature vector; (iii) a step of preprocessing the image feature vectors extracted from a training data set or the image feature vectors extracted before no crowd behavior class is built or when the crowd behavior class is not yet determined in the (ii) step; (iv) a step of generating situation networks expressing a subordinate relationship between the elements forming the image feature vectors extracted in the image section using a graph; (v) a step of extracting at least a route patterns from the situation networks for each of the crowd behavior class; and (vi) a step of setting the classification device in order for the crowd behavior class corresponding to be determined when the image feature vectors are input, based on the extracted route patterns.;COPYRIGHT KIPO 2016