Camera Calibration Using Pedestrian Height Deviation Minimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for automatically determining camera calibration parameters are either restrictive, expensive, or not universally applicable, as they often require manual annotation or complex vanishing point estimation.
Innovation Solution
A method using a convolutional neural network (CNN) to detect and select relevant elements of interest, such as people, and apply a geometric model with an optimization algorithm to determine calibration parameters like transverse inclination angle, focal length, and camera height, minimizing deviations in characteristic dimensions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual measurement methods are used to determine camera calibration parameters, then measurement precision can be achieved, but the method becomes time-consuming and expensive
Solution Approach 1:
The system uses randomly captured images containing pedestrians as automatic calibration subjects. The calibration process serves itself by utilizing naturally occurring elements in the scene rather than requiring dedicated calibration objects or manual intervention. Pedestrians in the images automatically provide the necessary reference information for determining camera parameters.
Solution Approach 2:
The patent replaces manual measurement mechanisms with an automated computational approach using optimization algorithms. Instead of physical measurement tools and human operators, the system uses mathematical optimization to automatically derive calibration parameters from image data, substituting mechanical/manual processes with computational ones.
2Productivity
If automatic calibration methods using pedestrian data are used, then calibration time is reduced, but the method becomes restrictive requiring specific pedestrian movements
Solution Approach 1:
The system uses more pedestrians than the minimum required for calibration, utilizing all available pedestrians in randomly captured images. This excessive use of calibration subjects provides redundancy and robustness, allowing calibration to proceed even when individual pedestrians have suboptimal poses or positions, thereby increasing adaptability.
Solution Approach 2:
The optimization algorithm iteratively adjusts calibration parameters to minimize the difference between measured and expected pedestrian heights. By changing parameters dynamically during optimization rather than requiring fixed predetermined conditions, the system adapts to various pedestrian configurations and image conditions.
3Extent of automation
If vanishing point estimation methods are used for calibration, then automation is achieved, but the computational complexity increases
Solution Approach 1:
The patent extracts only the essential information needed for calibration - pedestrian bounding boxes and their vertical positions - from the images. By extracting only the necessary data elements rather than performing full vanishing point estimation or comprehensive scene analysis, the system achieves automation with reduced computational complexity.
Solution Approach 2:
The calibration process is segmented into distinct steps: detecting pedestrians, extracting their vertical positions, computing expected heights from database averages, and optimizing parameters to minimize differences. This segmentation breaks down the complex calibration task into simpler, more manageable computational steps.
4Measurement precision
If manual annotation of calibration subjects is required, then measurement precision can be maintained, but the ease of operation deteriorates
Solution Approach 1:
The system automatically detects and utilizes pedestrians in the images without requiring manual annotation. The calibration process serves itself by automatically identifying calibration subjects and extracting necessary measurements, eliminating the need for human operators to manually mark or select pedestrians.
Solution Approach 2:
Manual annotation operations are replaced with automated computer vision algorithms. Instead of human operators manually marking pedestrians, the system uses automated detection and measurement algorithms to extract pedestrian positions and dimensions, substituting manual mechanical annotation with computational automation.
Data Source
Figure 1
Figure 2
Figure 3(a0)~3(f)
AI summary
The present invention relates to a method for determining calibration parameters of a camera, comprising the steps of: (a) Obtaining at least one image acquired by said camera representing a plurality of features of interest of a given category; (b) Detecting said features of interest in said image; (c) Selecting a subset of the detected features of interest; (d) For each feature of interest in said subset, expressing a characteristic dimension of said feature of interest as a function of the representation of said feature of interest in the image and the calibration parameters by means of a geometric model; (e) Determining, using an optimization algorithm, the value of the calibration parameters minimizing a difference between the expressions of said characteristic dimension and an expected average value of said characteristic dimension.The present invention also relates to a method for calibrating a camera and a method for measuring a real distance from an image acquired by a camera.