Automatic Field of View Error Detection System Based on Camera Video Information
An image information-based system automatically detects and corrects camera field of view errors using YOLO and AKAZE algorithms, addressing inefficiencies and safety risks in vehicle and surveillance systems.
Patent Information
- Application Number
- KR1020260025907
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2026-02-09
- Publication Date
- 2026-11-11
- Estimated Expiration
- 2046-02-09
AI Technical Summary
Existing camera systems in vehicles and surveillance environments suffer from field of view errors due to environmental factors and structural issues, requiring manual adjustment and leading to operational inefficiencies and potential safety risks.
An image information-based system that automatically detects field of view errors by comparing reference and current images using YOLO-based object removal, AKAZE feature point extraction, BFMatcher matching, and RANSAC-based Homography calculation to quantify positional discrepancies.
Enables automatic detection and correction of camera misalignments, improving operational efficiency and safety by preventing system failures and reducing manual intervention in large-scale surveillance systems.
Smart Images

Figure 112026017063560-PAT00003_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to an image information-based automatic camera field of view error detection system and method.
[0002] Specifically, the invention relates to an image information-based camera field of view error automatic detection system and method capable of automatically detecting camera field of view (Pan / Tilt / Zoom) errors or physical distortion by comparing a reference image captured by a camera with a current image and through feature point-based image registration and common field of view (overlap) area analysis.
[0003] To this end, the present invention performs YOLO-based object removal, AKAZE feature point extraction, BFMatcher matching, RANSAC-based Homography calculation, etc., and analyzes the common region of the two calculated images to quantify the field of view position error. Background Technology
[0005] Recently, cameras that capture images of the front, rear, or sides are being installed and used in vehicles to provide convenience for the driver during driving and parking.
[0006] As the vehicle is continuously used after delivery, the tolerances corrected at the time of delivery change due to various environmental factors such as vehicle vibration, folding of side mirrors, and opening and closing of vehicle doors, causing errors to gradually occur in the images displayed on the screen.
[0008] As a result, drivers experience the inconvenience of having to use distorted images when driving or parking.
[0009] To resolve this, the changed tolerances must be corrected; however, currently, there is the inconvenience of having to visit a service center or business office capable of tolerance correction for repairs.
[0011] Meanwhile, to resolve these problems, Korean Patent Publication No. 10-2011-0126178 (Title of Invention: Vehicle Rear Camera Control System and Method) discloses a technology that enables the rear camera to operate continuously according to the driver's operation, even when reversing, parking, or driving.
[0012] In other words, in the case of the prior art, the driver views the image and manually operates the angle adjuster to generate an angle adjustment setting signal to operate the rear camera, and the angle of the camera can be adjusted by the angle adjustment motor drive unit.
[0013] However, with this conventional technology, the driver must manually operate an angle adjuster to adjust the camera angle, which entails inconvenience for the driver to adjust directly.
[0014] In addition, since it must be adjusted manually, the angle is set differently each time it is operated, causing a sense of incongruity in the image output to the display unit.
[0016] As an attempt to solve this problem, Registered Patent Publication No. 10-2543523 describes a camera error correction system and method.
[0017] The above technology describes an error correction system for a camera in a Camera Monitor System (CMS), comprising one or more cameras attached to each side of a vehicle, a memory storing a program for correcting errors in an image captured by said cameras, and a processor for executing the program stored in said memory. By executing said program, said processor detects a correction point from an image captured by said cameras, generates a correction parameter based on a result of comparing the detected correction point with a preset correction point, and corrects the error of said captured image based on said correction parameter.
[0019] Regarding such camera calibration, there are issues with the structural problems of PTZ cameras; problems with physical misalignment due to the installation environment; and problems with manual-based post-treatment.
[0020] In particular, structural issues with PTZ cameras include reduced positional accuracy due to decreased durability of internal motors and gears during long-term operation, and discrepancies between the set field of view and the actual field of view resulting from the accumulation of Pan / Tilt driving errors during continuous touring (preset cycling).
[0021] In addition, regarding the issue of physical misalignment due to the installation environment, outdoor cameras are directly affected by external factors such as wind, impact, and vibration, which can cause the camera direction to become physically misaligned. Furthermore, there is a risk that the field of view may gradually change or suddenly become misaligned due to bracket loosening, equipment aging, etc.
[0022] In particular, because only manual, reactive responses are possible, camera field of view misalignment is typically detected only after a failure occurs, requiring monitoring personnel to inspect the site in person. Furthermore, there is a problem where operational efficiency is reduced as corrective measures are taken only after system errors or video analysis failures. Prior art literature
[0024] Published Patent Application No. 10-2011-0126178 Registered Patent Application No. 10-2543523 Registered Patent Application No. 10-1348800 Registered Patent Application No. 10-2479253 The problem to be solved
[0025] The objective of the present invention is to provide an image information-based camera field of view error automatic detection system and method capable of automatically detecting camera field of view (Pan / Tilt / Zoom) errors or physical distortion by comparing a reference image captured by a camera with a current image, through feature point-based image matching and common field of view (overlap) area analysis.
[0027] To this end, the present invention provides an image information-based camera field of view error automatic detection system and method that performs YOLO-based object removal, AKAZE feature point extraction, BFMatcher matching, RANSAC-based Homography calculation, etc., and analyzes the common area of two calculated images to quantify the field of view position error. means of solving the problem
[0029] The image information-based camera field of view error automatic detection system and method according to the present invention, devised to achieve the above-described purpose, is an image information-based camera field of view error automatic detection system and method that automatically detects a camera field of view error based on current image information input in real time and reference image information already secured.
[0030] An input unit that performs the function of receiving and inputting image information from a camera;
[0031] An object removal unit that removes elements interfering with image alignment on image information input through the above input unit using YOLO Segmentation;
[0032] A feature point extraction unit that performs the function of extracting feature points using the AKAZE algorithm in a state where a specific object is separated and removed through an object removal unit from image information input through the above input unit;
[0033] A feature point matching and matching unit that matches feature points extracted through the above feature point extraction unit using a BFMatcher and aligns them through RANSAC-based Homography calculation;
[0034] A common field of view area calculation unit that calculates a common field of view area using the Overlap Bounding Box technique to calculate a common field of view between the above-mentioned current image information and reference image information;
[0035] It is characterized by including a field of view position error determination unit that analyzes the difference in the common area between the current image information and the reference image information according to the calculation result of the common field of view area calculation unit, and automatically reports that the camera is misaligned if the analysis result exceeds a threshold.
[0037] At this time, the difference for the common area analyzed by the above-mentioned field of view position error determination unit is characterized by being a difference for one or more selected from the amount of change in movement, rotation, and enlargement / reduction of the above-mentioned common area. Effects of the invention
[0039] According to the image information-based camera field of view error automatic detection system and method of the present invention, YOLO-based object removal, AKAZE feature point extraction, BFMatcher matching, RANSAC-based Homography calculation, etc. are performed, and the common area of the two calculated images is analyzed to quantify the field of view position error.
[0040] Accordingly, the present invention enables the automatic detection of camera field of view (Pan / Tilt / Zoom) errors or physical distortion by comparing a reference image captured by a camera with a current image, through feature point-based image matching and common field of view (overlap) area analysis.
[0042] According to the image information-based automatic camera field of view error detection system and method of the present invention, it is possible to automatically check whether a city security camera is misaligned due to external impact and environmental influences, prevent in advance situations where control becomes impossible due to a change in the field of view, and enable periodic automatic inspection in large-scale CCTV operating environments, and it can be used in public CCTV and security control systems.
[0044] As another example, by using it in a traffic control system, changes in direction and offsets of road surveillance cameras (including PTZ) can be automatically detected, and problems such as traffic analysis / illegal parking analysis systems malfunctioning due to visibility errors can be prevented.
[0046] As another example, by using it in disaster / wildfire monitoring systems, it is possible to immediately detect cases where the field of view of wildfire detection CCTVs is distorted by wind or vibration, and it is possible to improve the reliability of remote unmanned monitoring systems.
[0047] By using it for industrial / facility monitoring, it is possible to automatically detect misalignment of surveillance cameras in facilities such as factories, logistics warehouses, and power plants, thereby preventing accidents related to facility safety monitoring, and increasing the speed of immediate alarms and maintenance requests when camera misalignment occurs.
[0048] Furthermore, in unmanned control / AI analysis systems, it can be utilized as an essential element in unmanned control centers / robot control platforms for maintaining input image quality of AI object recognition / behavior analysis systems and preventing AI false detection / non-detection caused by camera field of view errors, and
[0049] In large-scale CCTV operating environments (smart cities, government offices, etc.), it offers the advantages of enabling periodic automatic visibility checks for hundreds or thousands of cameras, reducing operating costs, maximizing management efficiency, and interoperability with existing Video Management Systems (VMS). Brief explanation of the drawing
[0051] FIG. 1 illustrates an image information-based camera field of view error automatic detection system and method according to the present invention. FIG. 2 is a flowchart illustrating the operation method of the image information-based camera field of view error automatic detection system and method according to the present invention. Specific details for implementing the invention
[0052] Terms and words used in this specification and claims should not be interpreted as being limited to their ordinary or dictionary meanings, but should be interpreted in a meaning and concept consistent with the technical spirit of the invention, based on the principle that the inventor can appropriately define the concept of the terms to best describe his invention.
[0054] Therefore, the embodiments described in this specification and the configurations illustrated in the drawings are merely the most preferred embodiments of the present invention and do not represent all of the technical ideas of the present invention; thus, it should be understood that various equivalents and modifications that can replace them may exist at the time of filing this application.
[0056] Before proceeding with the following description with reference to the drawings, it should be noted that matters not necessary to reveal the gist of the invention, namely known configurations that a person skilled in the art with ordinary knowledge can obviously add, have not been illustrated or specifically described.
[0058] The present invention relates to an image information-based automatic camera field of view error detection system and method.
[0059] Specifically, the invention relates to an image information-based camera field of view error automatic detection system and method capable of automatically detecting camera field of view (Pan / Tilt / Zoom) errors or physical distortion by comparing a reference image captured by a camera with a current image and through feature point-based image registration and common field of view (overlap) area analysis.
[0060] To this end, the present invention performs YOLO-based object removal, AKAZE feature point extraction, BFMatcher matching, RANSAC-based Homography calculation, etc., and analyzes the common region of the two calculated images to quantify the field of view position error.
[0062] The image information-based camera field of view error automatic detection system and method according to the present invention will be explained with reference to the attached drawings.
[0063] FIG. 1 shows an image information-based camera field of view error automatic detection system and method according to the present invention, and FIG. 2 shows a flowchart of the operation method of the image information-based camera field of view error automatic detection system and method according to the present invention.
[0065] The image information-based camera field of view error automatic detection system and method according to Fig. 1 of the attached drawings comprises a platform including an input unit; an object removal unit; a feature point extraction unit; a feature point matching and alignment unit; a common field of view area calculation unit; and a field of view position error determination unit.
[0066] Such a platform consists of software or hardware that can be accessed via the Web or an App, and may be implemented as a physical or virtual server. It does not matter how it is designed, as long as it is obvious to a person skilled in the art at the time of filing the application for this invention.
[0068] The above-mentioned input unit performs the function of receiving image information from the camera. At this time, the image received in real time through the input unit is referred to as the current image information, and the image used for comparison with the current image information is referred to as the reference image information. That is, the reference image information is image information input in a normal state without camera distortion, and serves as the object of comparison with the current image information, acting as a standard for determining the error regarding the camera's field of view based on the current image information.
[0070] In addition, the object removal unit performs the function of removing elements that interfere with image alignment on the image information input through the input unit, and to this end, it removes moving objects (vehicles / people) using YOLO Segmentation.
[0071] To elaborate, the YOLO series is a representative model that started with object detection and has now expanded to segmentation.
[0072] Among these YOLO series, the YOLOv11-Seg Full Architecture, in particular, separates (removes) specific objects through the process from Backbone → Neck → Detection Head → Segmentation Head (see [Table 1]).
[0074]
[0075] The Backbone refers to the role of receiving input images and extracting features while gradually reducing the resolution. Additionally, the Neck creates multi-scale features by combining FPN and PAN structures, the Detection Head predicts classes and bounding boxes from each scale feature, and YOLO uniquely uses the Proto + Coeff method for segmentation.
[0077] The above feature point extraction unit performs the function of extracting feature points from image information input through the input unit, in a state where a specific object is separated and removed through the object removal unit. That is, it performs the function of extracting only the feature points of the filtered background (AKAZE) and extracting MLDB descriptors.
[0078] The above AKAZE refers to a feature point (keypoint) detection and descriptor algorithm that accelerates the non-linear scale space of KAZE.
[0079] For further explanation regarding feature extraction using the AKAZE algorithm, please refer to https: / / www.koreascience.or.kr / article / JAKO202426443231914.pdf.
[0081] The feature point matching and alignment unit performs the function of matching and aligning feature points extracted through the feature point extraction unit. Specifically, the extracted feature points are matched through BFMatcher, and alignment is performed through RANSAC-based Homography calculation.
[0082] To elaborate on BFMatcher, it calculates and matches pairs of coordinates representing the same point in two image information sets based on the current image information input through the input section and the reference image, based on [Table 2].
[0083] https: / / blog.naver.com / qkrdnjsrl0628 / 223755246113
[0085]
[0087] RANSAC-based Homography is a method for robustly estimating the homography matrix (H) when there are outliers mixed in the feature point correspondence pairs.
[0089] The above common field of view calculation unit calculates an overlap bounding box using the projection result of a reference image in order to calculate the common field of view (overlap region) of the two images.
[0090] At this time, the Overlap Bounding Box technique refers to a technique that calculates how much two (or more) rectangular bounding boxes (bounding boxes) overlap each other (intersection area ratio) or removes duplicates when there is a lot of overlap to determine the final selected box.
[0091] In addition, the above-mentioned field of view position error determination unit analyzes the amount of change in movement, rotation, and magnification / reduction of the common area between the reference image and the current image information based on the calculation result of the above-mentioned common field of view area calculation unit, and automatically reports that the camera is misaligned if the analysis result exceeds a threshold.
[0092] Since the comparison between two images using this Overlap Bounding Box technique is described in known technology for determining the degree of freedom of an object, specific details will be omitted.
[0094] According to the image information-based camera field of view error automatic detection system and method of the present invention configured as described above, YOLO-based object removal, AKAZE feature point extraction, BFMatcher matching, RANSAC-based Homography calculation, etc. are performed, and the common area of the two calculated images is analyzed to quantify the field of view position error.
[0095] Accordingly, the present invention enables the automatic detection of camera field of view (Pan / Tilt / Zoom) errors or physical distortion by comparing a reference image captured by a camera with a current image, through feature point-based image matching and common field of view (overlap) area analysis.
[0097] According to the image information-based automatic camera field of view error detection system and method of the present invention, it is possible to automatically check whether a city security camera is misaligned due to external impact and environmental influences, prevent in advance situations where control becomes impossible due to a change in the field of view, and enable periodic automatic inspection in large-scale CCTV operating environments, and it can be used in public CCTV and security control systems.
[0099] As another example, by using it in a traffic control system, changes in direction and offsets of road surveillance cameras (including PTZ) can be automatically detected, and problems such as traffic analysis / illegal parking analysis systems malfunctioning due to visibility errors can be prevented.
[0101] As another example, by using it in disaster / wildfire monitoring systems, it is possible to immediately detect cases where the field of view of wildfire detection CCTVs is distorted by wind or vibration, and it is possible to improve the reliability of remote unmanned monitoring systems.
[0102] By using it for industrial / facility monitoring, it is possible to automatically detect misalignment of surveillance cameras in facilities such as factories, logistics warehouses, and power plants, thereby preventing accidents related to facility safety monitoring, and increasing the speed of immediate alarms and maintenance requests when camera misalignment occurs.
[0103] Furthermore, in unmanned control / AI analysis systems, it can be utilized as an essential element in unmanned control centers / robot control platforms for maintaining input image quality of AI object recognition / behavior analysis systems and preventing AI false detection / non-detection caused by camera field of view errors, and
[0104] In large-scale CCTV operating environments (smart cities, government offices, etc.), it offers the advantages of enabling periodic automatic visibility checks for hundreds or thousands of cameras, reducing operating costs, maximizing management efficiency, and interoperability with existing Video Management Systems (VMS).
[0106] The description above using the drawings describes only the main aspects of the present invention, and it is obvious that the present invention is not limited to the configuration of the drawings, as various designs are possible within the technical scope.
Claims
Claim 1 Current video information entered in real time and; An image information-based automatic camera field of view error detection system that automatically detects a camera field of view error based on previously acquired reference image information; comprising: an input unit that performs the function of receiving image information from a camera; an object removal unit that removes elements hindering image alignment on the image information input through the input unit using YOLO Segmentation; a feature point extraction unit that performs the function of extracting feature points using the AKAZE algorithm when a specific object is separated and removed from the image information input through the input unit using the object removal unit; a feature point matching and alignment unit that matches the feature points extracted through the feature point extraction unit using BFMatcher and aligns them using RANSAC-based Homography calculation; a common field of view area calculation unit that calculates a common field of view area using the Overlap Bounding Box technique to calculate a common field of view between the current image information and the reference image information, and calculates the common field of view (overlap region) between the current image information and the reference image information based on the intersection area ratio; and, according to the calculation result of the common field of view area calculation unit, in the common area between the current image information and the reference image information An image information-based camera field of view error automatic detection system comprising: a field of view position error determination unit that analyzes differences and automatically reports that the camera is misaligned if the analysis result exceeds a threshold; wherein the difference for a common area analyzed by the field of view position error determination unit is a difference for one or more selected from the amount of change in movement, rotation, and enlargement / reduction of the common area. Claim 2 delete
Citation Information
Patent Citations
Dynamic scene SLAM method based on YOLO algorithm and GMS feature matching
CN111161318A
Method and system for dynamically calibrating vehicular cameras
US20150049193A1
