Vision-Aided Inertial Navigation Loop Closure
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Visual odometry systems in navigation accumulate errors over time due to wheel slip and camera resolution limitations, and loop closure detection is challenging without precise six-degree-of-freedom transform knowledge.
Innovation Solution
A vision-aided inertial navigation system incorporating a multi-state constraint Kalman filter (MSCKF) with a loop closure detector using convolutional neural networks to classify images and reset navigation solutions, incorporating a strapdown integrator for inertial data and a pose graph solver to distribute error across six degrees of freedom.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If visual odometry is used to estimate position and orientation from image features, then navigation solution can be obtained without direct distance measurement sensors, but accumulated error increases over time due to wheel slip and camera resolution limitations
Solution Approach 1:
The patent combines visual odometry with inertial measurement unit (IMU) data in a vision-aided inertial navigation system. The IMU provides complementary information about vehicle motion that does not suffer from the same accumulation errors as visual odometry alone, particularly for wheel slip compensation. This fusion of multiple sensing modalities resolves the contradiction by maintaining navigation versatility while improving position accuracy through redundant measurement sources.
Solution Approach 2:
The system implements loop closure detection that compares current images with stored historical images to identify when the vehicle returns to a previously visited location. When loop closure is detected, the system uses the known transform between the loop closure image and matching image to correct accumulated navigation errors. This feedback mechanism continuously resets and corrects the navigation solution, preventing error accumulation while maintaining ongoing navigation capability.
2Measurement precision
If loop closure detection is implemented to correct accumulated errors, then navigation accuracy improves when returning to previous locations, but the transform between camera poses is not known in all six degrees of freedom
Solution Approach 1:
The patent introduces an inertial measurement unit (IMU) as an intermediary that provides the missing degree of freedom information during loop closure. The IMU data fills the gap in the six-degree-of-freedom transform by providing independent measurements of vehicle orientation and position changes that are not derivable from image analysis alone. This intermediary sensor enables complete pose reconstruction when loop closure is detected, resolving the information loss problem.
Solution Approach 2:
The system design makes the navigation system multi-functional by integrating both visual odometry and inertial navigation capabilities. The same sensor fusion framework handles both continuous navigation estimation and loop closure correction, with the IMU serving dual purposes: providing ongoing motion compensation and supplying missing pose information during loop closure events. This universal approach resolves the six-degree-of-freedom transform problem by making the system self-sufficient across different operational modes.
3Reliability
If stored images are compared with incoming images for loop closure detection, then accumulated error can be cancelled, but false positives and negatives occur without strong comparison metrics
Solution Approach 1:
The system performs loop closure detection on a subset of images rather than every single image. Specifically, it compares incoming images with stored images at selected intervals or under specific conditions (e.g., when the vehicle is stationary or moving slowly). This partial action approach reduces the total number of comparisons needed, lowering computational complexity and reducing the likelihood of false positives, while still achieving sufficient error correction reliability through strategic sampling of the image sequence.
Data Source
AI summary
A vision-aided inertial navigation system determines navigation solutions for a traveling vehicle. A navigation solution module analyzes image sensor poses and estimated inertial navigation solutions to produce a time sequence of system navigation solution outputs that represent changing locations of the traveling vehicle. A loop closure detector performs a two-threshold image classification to classify an incoming navigation image as novel, loop-closing, or neither, and the navigation solution module resets the navigation solution output to represent a previous location when the loop closure detector classifies an incoming navigation image as loop-closing.