Person clothing feature extraction device, person search device, and processing method thereof

摘要

A person's region is detected from input video of a surveillance camera; a person's direction in the person's region is determined; the separability of person's clothes is determined to generate clothing segment separation information; furthermore, clothing features representing visual features of person's clothes in the person's region are extracted in consideration of the person's direction and the clothing segment separation information. The person's direction is determined based on a person's face direction, person's motion, and clothing symmetry. The clothing segment separation information is generated based on analysis information regarding a geometrical shape of the person's region and visual segment information representing person's clothing segments which are visible based on the person's region and background prior information. A person is searched out based on a result of matching between a clothing query text, representing a type and a color of person's clothes, and the extracted person's clothing features.

主权项

1. A person clothing feature extraction device using a processor and a clothing feature storage configured to store clothing features based on an input video, the person clothing feature extraction device comprising:
a person region detection part configured to automatically detect a person's region from the input video by performing a person detection image processing operation; a person direction determination part configured to automatically determine a person's direction of the person's region by performing an image processing operation; a clothing segment separation part configured to automatically determine a separability of person's clothes in the person's region based on shape analysis of the person's region and the position of the person's region using background prior information indicating a part of a person is visible, and configured to automatically produce clothing segment separation information including information whether or not the person's region can be separated into segment regions and information describing segment regions from which clothing features are to be extracted; and a clothing feature extraction part configured to extract the clothing features representing visual features of person's clothes from the segment regions of the person's region defined by the clothing segment separation information as clothing features corresponding to the person's direction, thus storing the clothing features in the clothing feature storage.