Dense feature point matching-based dome camera preset position offset automatic identification method

Abstract

The invention discloses a dense feature point matching-based dome camera preset position offset automatic identification method, and relates to the technical field of intelligent video monitoring. Themethod comprises the following steps: obtaining a preset position image and a to-be-detected image; extracting feature descriptors of the preset position image and the to-be-detected image respectively; establishing an optimization objective function; screening the feature point pairs; according to the screened feature point pairs, obtaining a homography matrix of the feature point pairs, and obtaining bias parameters in the x direction and the y direction; comparing the deviation degrees of the to-be-detected image in the two directions, and then judging whether the deviation of the to-be-detected image in the direction is greater than a confidence range set by a user or not; and if the offset is greater than the confidence range set by the user, returning the information that the to-be-detected image generates an offset and calculating the offset degree, otherwise, returning the information that the to-be-detected image does not generate the offset. According to the method, the deviated image is automatically corrected on the basis of dense feature point matching, so that the accuracy is ensured while the hardware cost is saved.

Description

technical field [0001] The invention relates to the technical field of intelligent video monitoring, in particular to a method for automatically identifying the offset of a dome machine preset position based on dense feature point matching. Background technique [0002] With the development of computer vision technology, intelligent video surveillance technology is widely used in scenes such as banks, companies, airports, customs and railway stations. When doing intelligent analysis of video surveillance for some specific scenes, the fixed preset position will be limited As a template. However, because the monitoring lens will change its shooting angle due to some accidental factors such as external force or occlusion, the preset position set in the scene will be biased. In actual scenes, the deviation of the preset position often occurs The shift is very small, but the results of the intelligent analysis of the surveillance video are very different. In order to timely and a...

AI Technical Summary

Problems solved by technology

[0002] With the development of computer vision technology, intelligent video surveillance technology is widely used in scenes such as banks, companies, airports, customs and railway stations. When doing intelligent analysis of video surveillance for some specific scenes, the fixed preset position will be limited As a template. However, because the monitoring lens will change its shooting angle due to some accidental factors such as external force or occlusion, the preset position set in the scene will be biased. In actual scenes, the deviation of the preset position often occurs The shift is very small, but the results of the intelligent analysis of the surveillance video are very different. In order to timely and accurately find the offset of the preset position and complete the automatic correction of the picture before the intelligent processing of the video, a ball based on dense feature point matching is proposed. Machine preset offset automatic identification method.

[0003] The current pre-position offset detection methods mainly include the following: background modeling and template matching. Judging whether the preset position is offset, but the change of the pixel value is not very reliable. When the change of the pixel value is greatly affected by the lighting conditions, the probability of abnormal detection of the preset position offset is very high. The template matching method is Sliding window matching is performed in the image through a pre-established template. When the similarity of a certain area is higher than the set threshold, the matching is considered successful. The accuracy of this method depends heavily on the selection of the template, and it is not true in the actual scene. Very practical

In order not to affect the results of intelligent video analysis, the invention patent with the publication number CN105812720 "a method and device for correcting the preset position of a dome camera" and the invention patent with the authorized announcement number CN102937816B "a method and method for calibrating the deviation of the camera preset position" The correction of the offset position of the "device" adopts hardware adjustment to adjust the offset problem of the preset position, but adding hardware to solve this type of problem is costly and there are certain errors in the adjustment

Images