Visual Odometry Mistracked Point Detection via Back-Projection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional techniques for processing image sequences to determine the position of moving objects suffer from increasing errors due to inaccuracies in tracking feature points between frames, particularly in environments where GPS data is unreliable or unavailable.

Innovation Solution

A moving image processing device that includes units for obtaining moving images, extracting feature points, tracking their matching relationships, calculating real-space coordinates, back-projecting these coordinates, and detecting mistracked points using different methods and conditions to improve tracking accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If feature point tracking is performed continuously between frames to obtain position information, then position detection capability is maintained, but tracking accuracy deteriorates due to accumulated errors

Engineering Contradiction:

Improveposition detection capabilityVSAvoidtracking accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent implements feedback by back-projecting calculated three-dimensional coordinates of feature points onto image planes and comparing these back-projected coordinates with actually detected coordinates. This comparison generates error information that feeds back into the tracking process, enabling detection and correction of mistracked points to maintain accuracy despite continuous processing

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent replaces purely mechanical/algorithmic tracking with a hybrid approach incorporating geometric verification through back-projection. By substituting direct tracking with a verification mechanism that uses coordinate transformation and comparison, the system can identify and correct tracking errors without compromising continuous position detection

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Reliability

If multiple methods are used to calculate back-projected coordinates for detection, then detection reliability improves, but device complexity increases

Engineering Contradiction:

Improvedetection reliabilityVSAvoidprocessing complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the detection process into distinct components: feature point extraction, three-dimensional coordinate calculation, back-projection to image plane, and coordinate comparison. This segmentation allows multiple methods to be applied to specific segments (such as different back-projection methods) without overwhelming complexity, as each segment can be optimized independently

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS9171225B2Device, method, and recording medium for detecting and removing mistracked points in visual odometry systems

Publication Date: 2015.10.27 TOPCON CORPORATION

AI summary

Three-dimensional coordinates of feature points of an object to be measured are back-projected to a frame image photographed from a specific position, and image coordinates of the back-projected feature points and the feature points in this frame image are compared. In this case, the feature points, which are mismatched, are removed as feature points which are mistracked between plural frames. In this case, two processing systems, of which initial conditions of calculation for obtaining the back-projected coordinates are different from each other, are performed, and the detection of the above mistracked points is performed on each of the two back-projected coordinates. The mistracked points detected in at least one of the processing systems are removed, and are not succeeded to the following processing.