Camera Roll Angle Estimation Using Ground Plane Geometry
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Advanced Driver Assisted Systems (ADAS) struggle to accurately calibrate and determine the roll angle of cameras mounted on vehicles, especially trucks, due to changes in load and suspension, which can lead to reduced performance and errors in lane detection, as prior methods assume a negligible roll angle or require computationally expensive Optical Flow calculations.
Innovation Solution
A method to estimate the roll angle of a camera using a simple formula based on Cartesian axes transformation, Inverse Perspective Mapping, and Taylor expansion, which calculates the roll angle from the coordinates of two points on the ground plane, such as lane markers, without requiring high processing power.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Optical Flow based method is used for roll angle estimation, then measurement precision is improved, but use of energy and computational complexity increase
Solution Approach 1:
The patent extracts only the essential geometric relationships needed for roll angle calculation from the full Optical Flow method. By focusing specifically on the transformation between camera coordinate system and ground plane coordinate system, the solution removes unnecessary computational complexity while retaining the core functionality of roll angle estimation.
Solution Approach 2:
The patent uses a simplified mathematical model that copies the essential geometric transformation principles from Optical Flow but implements them through direct coordinate transformation formulas rather than full Optical Flow computation. This allows achieving similar measurement precision with significantly reduced computational requirements.
2Device complexity
If constant roll angle assumption is made in auto-calibration, then device complexity is reduced, but measurement precision deteriorates
Solution Approach 1:
The patent transitions from a static constant roll angle assumption to a dynamic roll angle determination system. The roll angle is now calculated dynamically based on real-time detection of ground plane features and coordinate transformations, allowing the system to adapt to changing vehicle conditions while maintaining relatively simple implementation through mathematical formulas.
3Ease of operation
If camera roll angle calibration is not performed, then ease of operation is improved, but lane detection precision deteriorates
Solution Approach 1:
The patent implements a self-calibrating system that automatically determines the roll angle using visible ground plane features such as lane markings. The system performs self-correction by detecting the geometric relationships in the captured image and calculating the appropriate roll angle compensation, eliminating the need for manual calibration operations while ensuring accurate lane detection.
Data Source
AI summary
A method of determining the roll angle of a camera mounted on a host vehicle includes:a) defining Cartesian axes in the real world and corresponding axes in the camera imager;b) from an image taken from the camera, determining the co-ordinates of the two points (P1, P2) of the real world, in the imager of the camera ((x I,1, y I,1), (x I,2, y I,2));c) determining the difference (D) in the coordinates of the points in a horizontal World plane in one axis (Xw,1−Xw,2), using corresponding co-ordinate system as in step b); andd) determining the roll angle ρ of a camera located on a host vehicle from the following formulaρ=[D/h−(c2/a2−c1/a1)]/[(a2d2−b2c2)/a22−(a1d1−b1c1)/a12],where a1 and a2 are found from the following, where i=1,2 represents the two points (P1, P2):ai=sin θ*f+cos θ*yI,i;bi=−cos θ*xI,i ci=−cos η cos θ*f+sin η*xI,i+cos η sin θ*yI,I di=−cos η sin θ*xI,i+sin η*yI,i where for θ is the known value of fixed tilt angle, η is the known value of fixed pan angle and h is the known fixed camera height h; and xI,i, yI,i are the respective co-ordinates of two of the points in the imager.