KR-20260063752-A - Method and system for image-based vehicle position estimation robust to camera lens distortion by applying multi-homograph transformation algorithm

KR20260063752AKR 20260063752 AKR20260063752 AKR 20260063752AKR-20260063752-A

Abstract

A video-based vehicle location estimation method according to one embodiment of the present invention comprises: a step of acquiring video data within an indoor GPS blind spot; a step of detecting an object within the acquired video data and generating a grid map based on the detection result; and a step of estimating the object location by applying a multiple homograph matrix algorithm to the detected object based on the generated grid map. Accordingly, the vehicle location can be estimated in real time with less computational power than existing methods for CCTV footage already installed in indoor areas, underground parking lots, etc., where GPS reception strength is weak, without additional cost investment.

Inventors

장수현
신대교
장준혁
장성현
안병만

Assignees

한국전자기술연구원

Dates

Publication Date: 20260507
Application Date: 20241031

Claims (12)

A step in which the system acquires image data within an indoor GPS blind spot; The system detects objects within acquired image data and generates a grid map based on the detection results; and An image-based vehicle position estimation method comprising the step of estimating the object location by applying a multi-homograph matrix algorithm to detected objects based on a generated grid map.
In claim 1, The step of generating a grid map is, An image-based vehicle location estimation method characterized by detecting vehicles and pedestrians within acquired image data using a Convolutional Neural Network (CNN) and generating a grid map by reflecting the detected vehicles and pedestrians.
In claim 1, The step of estimating the object location is, An image-based vehicle position estimation method characterized by generating multiple homograph matrices through multiple corresponding points for a detected object, and estimating the object position by assigning weights to each of the generated multiple homograph matrices.
In claim 3, The step of estimating the object location is, An image-based vehicle position estimation method characterized by generating multiple homograph matrices using six or more corresponding points when generating multiple homograph matrices.
In claim 4, The step of estimating the object location is, An image-based vehicle position estimation method characterized by assigning weights to multiple homograph matrices, performing learning on the multiple homograph matrices with assigned weights to generate a single integrated homograph matrix, and applying the single integrated homograph matrix to estimate the object position.
In claim 5, The step of estimating the object location is, An image-based vehicle position estimation method characterized by assigning weights to each of multiple homograph matrices, and then learning the weights assigned to each homograph matrix through a Multi-Layer Perceptron (MLP) machine learning technique to finally generate a single integrated homograph matrix.
In claim 6, The step of estimating the object location is, Image-based vehicle position estimation method characterized by generating 9 homograph matrices using 6 corresponding points, wherein a single integrated homograph matrix is calculated through the following Equations 1 and 2 based on weights learned through an MLP machine learning technique for each of the 9 homograph matrices. (Formula 1) (Equation 2)
In claim 6, The step of estimating the object location is, An image-based vehicle position estimation method characterized by learning weights assigned to each homograph matrix through an MLP machine learning technique, such that the total loss between the actual value and the transformed value through the homograph matrix is minimized.
In claim 8, The total error (Loss) of the transformed values is, An image-based vehicle position estimation method characterized by being calculated through the following Equation 3 when the coordinates of the vehicle position predicted by multiple homograph projection transformation are (X i ,Y i ) and the coordinates of the actual vehicle position are X gt ,Y gt . (Equation 3)
A communication unit for acquiring video data within an indoor GPS blind spot; and An image-based vehicle position estimation system comprising: a processor that detects objects within acquired image data, generates a grid map based on the detection results, and estimates the object location by applying a multi-homograph matrix algorithm to the detected objects based on the generated grid map.
A step in which the system detects objects within image data and generates a grid map by reflecting the detected objects; The system generates a plurality of homograph matrices through a plurality of corresponding points for a detected object; The system assigns weights to each of the generated multiple homograph matrices; The system performs learning of a plurality of homograph matrices, each assigned weight, to generate a single integrated homograph matrix; and An image-based vehicle position estimation method comprising the step of estimating an object position by applying a single integrated homograph matrix to the system.
A grid map generation unit that detects objects within image data and generates a single grid map by reflecting the detected objects; A homograph matrix generation unit that generates multiple homograph matrices through multiple corresponding points for a detected object, assigns weights to each of the generated multiple homograph matrices, and performs learning on the multiple homograph matrices assigned weights to generate a single integrated homograph matrix; and An image-based vehicle position estimation system comprising: an object position estimation unit that estimates object positions by applying a single integrated homograph matrix.

Description

Method and system for image-based vehicle position estimation robust to camera lens distortion by applying multi-homograph transformation algorithm The present invention relates to a method and system for estimating a vehicle's location based on image analysis, and more specifically, to a method and system for estimating the location of a vehicle detected through the analysis of CCTV images installed in a parking lot. Recently, active technological development is underway for Auto Valet Parking services, a field of autonomous driving technology. To develop such autonomous parking technology, the development of technology to determine the location of a vehicle in real time from the moment it enters the parking lot must be prioritized. While GPS data can be used for outdoor parking lots, it has the disadvantage of being difficult to utilize due to large measurement errors in GPS blind spots, such as indoor parking lots. Accordingly, various indoor location tracking technologies are being developed, such as methods to estimate the location of detected vehicles by installing sensors like Bluetooth tags, beacons, RFID (Radio Frequency Identification), and UWB (Ultra-Wide Band) in GPS blind spots—like indoor parking lots with weak GPS reception—and methods to estimate the vehicle's location by generating a Point Cloud Data (PCD) map of the parking lot in advance and installing sensors like Lidar on the vehicle to perceive the surrounding environment and compare it with the PCD map; however, these technologies have limitations in that installing sensors requires significant time and cost. Conversely, while object location estimation methods based on camera image analysis using deep learning technology show reasonable accuracy, they have the disadvantage of requiring expensive analysis equipment due to the excessive amount of computation. Existing computer vision algorithms include projection transformation methods that align positional relationships on two different planes; however, due to camera lens characteristics, the image becomes distorted towards the edges, causing straight lines to appear curved. This has a disadvantage in that it limits accurate position estimation. To address the shortcomings of such projection transformations, there exists a camera lens calibration method that corrects images by determining the intrinsic and extrinsic variable values of the camera; however, this method has limitations in that applying it to all cameras in a parking lot is costly. FIG. 1 is a drawing provided for the description of the configuration of an image-based vehicle position estimation system according to an embodiment of the present invention, FIG. 2 is a drawing provided for a more detailed configuration description of the processor illustrated in FIG. 1. FIG. 3 is a diagram illustrating an example of how the location of an object detected on a grid map generated through an image-based vehicle location estimation system according to an embodiment of the present invention is represented. FIG. 4 is a drawing provided to explain the concept of homograph transformation, FIGS. 5 and 6 are drawings provided to explain the phenomenon of camera lens distortion. FIG. 7 is a drawing provided for explaining a plurality of homograph matrices generated through an image-based vehicle position estimation system according to an embodiment of the present invention. FIG. 8 is a diagram illustrating the learning of a plurality of homograph matrices through an image-based vehicle position estimation system according to an embodiment of the present invention. FIG. 9 is a drawing provided for explaining a loss function for generating a single integrated homograph matrix through an image-based vehicle position estimation system according to an embodiment of the present invention, and FIG. 10 is a flowchart provided to describe an image-based vehicle position estimation method according to one embodiment of the present invention. The present invention will be described in more detail below with reference to the drawings. To clearly explain the invention, parts unrelated to the description have been omitted from the drawings, and in the drawings, the width, length, thickness, etc., of the components may be exaggerated for convenience. FIG. 1 is a drawing provided to describe the configuration of an image-based vehicle position estimation system according to one embodiment of the present invention. The image-based vehicle location estimation system according to the present embodiment (hereinafter collectively referred to as the "system") is provided to estimate the location of an object detected through the analysis of CCTV images installed in an indoor GPS blind spot, such as an indoor parking lot. Referring to FIG. 1, the system may include a communication unit (100), a processor (200), and a storage unit (300). The communication unit (100) is equipped with a communication module connected to a network, and can receive and acquire vi