Stereoscopic Camera Calibration Using Optical Flow and Epipolar Lines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Classical stereoscopic vision systems with identical cameras on a single imaging plane face challenges in diversity of imaging devices and varying locations, leading to complex system installation and high computational loads during image analysis, particularly in advanced driver-assistance systems where cameras have different characteristics and locations.
Innovation Solution
A hybrid stereoscopic system using multiple, identical and/or non-identical cameras at initially unknown locations, where a processor calibrates the cameras by identifying optical flow and applying epipolar geometric constraints to converge epipolar lines, reducing computational costs and improving detection capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If classical stereoscopic systems use identical cameras on a single imaging plane, then mathematical complexity for object matching is reduced, but device complexity and installation difficulty increase when diverse cameras with different locations are used
Solution Approach 1:
The system performs preliminary calibration to determine epipolar geometry parameters and establish epipolar lines before actual object matching occurs. This pre-computation of geometric relationships enables the system to handle diverse camera configurations without increasing computational complexity during runtime, as the calibration data is stored and reused for matching operations
Solution Approach 2:
The patent introduces epipolar lines as an intermediary geometric construct that mediates between diverse camera positions and the object matching process. By projecting 3D scene points onto 2D epipolar lines in the image plane, the system creates a standardized reference framework that simplifies matching regardless of camera diversity or location variations
2Productivity
If cameras are fixed on a single imaging plane, then object matching computational load is reduced, but mutual field of view decreases when cameras are positioned at diverse locations
Solution Approach 1:
The patent transitions from constrained 2D same-plane camera positioning to 3D spatial camera arrangement by utilizing epipolar geometry in three-dimensional space. The epipolar constraint equation incorporates 3D camera positions and orientations, allowing cameras to be positioned at different locations while maintaining computational efficiency through geometric projection onto 2D epipolar lines
3Reliability
If diverse cameras with different characteristics are used, then detection capabilities improve, but computational load during image analysis increases
Solution Approach 1:
The system performs preliminary calibration to compute and store epipolar geometry parameters including fundamental matrices and epipolar line equations before actual object detection and matching. This pre-computation phase processes the diverse camera characteristics once during calibration, allowing subsequent detection operations to use these pre-computed parameters efficiently without repeatedly processing raw camera diversity data
Solution Approach 2:
The patent extracts and separates the geometric calibration parameters from the object matching process. By extracting epipolar geometry parameters (fundamental matrix, essential matrix, epipolar lines) as independent pre-computed data, the system removes the computational burden of handling diverse camera characteristics from the runtime matching process, leaving only simple pixel coordinate comparisons along predefined epipolar lines
Data Source
AI summary
A system and method of stereoscopic image processing by at least one processor may include receiving, from a first imaging device, having a first field of view (FOV), and located at a first, initially unknown position, a first image of a scene; receiving, from a second imaging device, having a second, different FOV, and located at a second, initially unknown position, a second image of the scene; calculating a plurality of flow lines in the first image, wherein each flow line represents an optical flow between a pixel of the first image and a corresponding pixel of the second image; and calibrating the imaging devices by determining at least one parameter of relative position between the first imaging device and second imaging device, based on the calculated flow lines.