Method for updating query information for tracing target object from multi-camera and multi-camera system performing the same

Patent metadata

Field	Value
Publication number	US-11853350-B2
Application number	US-202117537790-A
Country	US
Kind code	B2
Filing date	Nov 30, 2021
Priority date	Aug 24, 2021
Publication date	Dec 26, 2023
Grant date	Dec 26, 2023

Abstract

Embodiments relate to a method for updating query information for tracing a target object from multi-camera including receiving a query information update command including query information for tracing a target object from multi-camera, searching for at least one image displaying the target object among a plurality of images captured by the multiple cameras, and updating the query information of a query image based on the at least one found image, and a multi-camera system performing the same.

First claim

The invention claimed is:

1. A method for updating query information for tracing a target object, performed by a system including multiple cameras and an image analysis server, the method comprising:
receiving a query information update command including query information for tracing the target object from the multiple cameras, the query information including at least one of a query image displaying the target object, features of the query image extracted from the query image, or attributes of the target object;
searching for at least one image displaying the target object among a plurality of images captured by the multiple cameras;
updating the query information of the query image based on the at least one found image, the updating the query information comprising obtaining new features of the query image based on first features of the at least one found image extracted from the plurality of images and the features of the query image; and
obtaining new features of the query image by obtaining, as the new features of the query image, an output result calculated by applying input features to a fully connected layer pre-stored in the image analysis server, wherein the input features are a concatenation of the first features of the at least one found image and the features of the query image.

2. The method according to claim 1 , further comprising:
extracting the features of the query image from the query image, before searching for the at least one image or updating the query information, when the query information update command includes the query image and does not include the features of the query image.

3. The method according to claim 2 , wherein updating the query information comprises:
updating the query information based on the new features of the query image.

4. The method according to claim 3 , wherein obtaining the new features of the query image comprises:
calculating a representative value for the first features of the at least one found image and the features of the query image as the new features of the query image,
wherein the representative value is an average or maximum value of a feature value of the image and a feature value of the query image.

5. The method according to claim 3 , wherein obtaining the new features of the query image comprises:
directly obtaining, as the new features, at least some of remaining features except the features that match the features of the query image among the first features of the at least one found image.

6. The method according to claim 3 , wherein the query information further includes capture time information of the camera,
the method further comprises:
classifying the at least one found image based on the capture time information in the query information for each preset capture time slot, and
wherein obtaining the new features of the query image comprises generating the new features of the query image for each capture time slot to update the query information for each capture time slot.

7. The method according to claim 6 , wherein the capture time slot is set on the basis of a point having a relatively large change in background luminance for each capture time in a background luminance distribution in one day.

8. The method according to claim 3 , wherein the query information further includes specification information of the camera,
the method further comprises:
classifying the at least one found image based on the specification information in the query information for each specification of the camera, and
wherein obtaining the new features of the query image comprises generating the new features of the query image for each specification of the camera to update the query information for each specification.

9. The method according to claim 8 , wherein the specification information of the camera includes at least one of camera type, image format, pixel or resolution.

10. The method according to claim 1 , wherein searching for the at least one image displaying the target object comprises:
determining each image of which image features match the features of the query image as an image displaying the target object, for each of the plurality of images obtained from the multiple cameras.

11. The method according to claim 1 , wherein searching for the at least one image displaying the target object comprises:
determining an object in each image as the target object based on a matching level or a matching number between attributes of the corresponding object and the attributes of the target object, for each of the plurality of images obtained from the multiple cameras.

12. The method according to claim 1 , wherein the fully connected layer is pre-trained to increase an importance of unique features shared between an object having the input features and the target object.

13. A computer-readable recording medium having recorded a program for performing the method according to claim 1 .

14. A multi-camera system, comprising:
multiple cameras to capture an object to generate an image; and
an image analysis server to receive the images from the multiple cameras, the image analysis server is configured to:
receive a query information update command including query information for tracing a target object from the multiple cameras, lithe query information including at least one of a query image displaying the target object, features of the query image extracted from the query image, or attributes of the target object;
search for at least one image displaying the target object among the plurality of images captured by the multiple cameras;
update the query information of the query image based on the at least one found image, the update of the query information comprising obtaining new features of the query image based on first features of the at least one found image extracted from the plurality of images and the features of the query image; and
obtain new features of the query image by obtaining, as the new features of the query image, an output result calculated by applying input features to a fully connected layer pre-stored in the image analysis server, wherein the input features are a concatenation of the first features of the at least one found image and the features of the query image.

Assignees

• Korea Inst Sci & Tech

Inventors

• Kim Ig Jae
• Choi Heeseung
• Kim Haksub
• Chae Seungho
• Yang Yoonsik

Classifications

• G06F16/532 Primary

  Query formulation, e.g. graphical querying · CPC title

• G06F16/55

  Clustering; Classification · CPC title

• G06F16/5854

  using shape and object relationship · CPC title

• G06F18/21

  Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation · CPC title

• G06F18/24

  Classification techniques · CPC title