Apparatus and method for detecting object using PTZ camera

摘要

An apparatus for detecting an object includes a filter for filtering a current input image and a background model generated based on a previous input image, a homography matrix estimation unit for estimating a homography matrix between the current input image and the background model, an image alignment unit for converting the background model by applying the homography matrix to a filtered background model and aligning a converted background model and a filtered current input image, and a foreground/background detection unit for detecting a foreground by comparing corresponding pixels between the converted background model and the filtered current input image.

主权项

1. An apparatus for detecting an object, comprising a hardware processor to implement:
a foreground/background detector configured to detect a foreground by comparing corresponding pixel between a background model, generated based on a previous input image, and a current input image; and a background updater configured to update the background model based on time information of each pixel of the current input image; and a tracker configured to track the detected foreground in next input images, wherein the foreground/background detector is configured to classify a current pixel of the current input image into a foreground pixel or a background pixel by comparing pixel values of the current pixel and a pixel of the background model corresponding to the current pixel when the current pixel belongs to a detection area estimated according to a result of tracking the detected foreground for a predetermined time, wherein the foreground/background detector is configured to classify the current pixel of the current input image into the foreground pixel or the background pixel by comparing pixel values of the current pixel and comparison pixels of background model, the comparison pixels corresponding to the current pixel and at least one neighboring pixel of the current pixel in the background model, when the current pixel does not belong to the estimated detection area, and wherein the background updater is configured to update the background model by using a weight value for each pixel of the current input image, the weight value being a reciprocal number of an age value determined according to a time information of each pixel of the current input image.