Ego Motion Estimation Using Scene Flow Clustering for Dynamic Environments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ego motion estimation methods for devices like autonomous vehicles and robots are inaccurate in dynamic environments due to the assumption of a stationary environment, leading to poor pose estimation when moving objects are present.
Innovation Solution
The method generates a scene flow field for input images segmented into spaces, clusters these spaces based on similar scene flows, and estimates ego motion information using a probability vector map to distinguish between stationary and dynamic regions, thereby improving accuracy in dynamic environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If odometry is used to estimate ego motion, then the method is simple and computationally efficient, but the accuracy deteriorates in dynamic environments with moving objects
Solution Approach 1:
The patent segments the image into multiple spaces and clusters them based on scene flow characteristics. This segmentation allows the system to differentiate between stationary and dynamic regions, improving ego motion estimation accuracy by focusing on stationary regions while maintaining computational efficiency through localized processing.
Solution Approach 2:
The patent introduces dynamic clustering that adapts to changing environmental conditions. The clustering algorithm dynamically adjusts groupings of spaces based on scene flow patterns, enabling the system to handle dynamic environments with moving objects while maintaining accurate ego motion estimation.
2Reliability
If the environment is assumed to be stationary for odometry, then the computation is simplified, but the reliability deteriorates when moving objects are present
Solution Approach 1:
The patent divides the image into multiple spaces and clusters them based on scene flow similarity. This segmentation enables the system to identify and focus on stationary regions for reliable pose estimation while separating dynamic regions, thereby improving reliability without requiring complex global processing.
Solution Approach 2:
The patent applies scene flow analysis selectively to clustered spaces rather than processing the entire image uniformly. By focusing computational resources on representative spaces from each cluster, the system achieves reliable pose estimation with reduced processing complexity compared to full-scene analysis.
3Measurement precision
If scene flow analysis is performed for all spaces, then the accuracy of distinguishing stationary and dynamic regions improves, but the computation time increases
Solution Approach 1:
The patent segments the image into multiple spaces and clusters them based on scene flow characteristics. This segmentation allows the system to identify stationary regions with high accuracy by analyzing representative spaces from each cluster, rather than processing every space individually, thereby reducing computation time.
Solution Approach 2:
The patent merges similar spaces into clusters based on scene flow similarity. By combining multiple spaces into representative clusters, the system achieves accurate stationary region identification through analysis of fewer clustered units, significantly reducing computation time while maintaining precision.
Data Source
AI summary
Disclosed is an ego motion estimation method and apparatus, wherein the apparatus calculates a scene flow field from a plurality of spaces of an input image, clusters the plurality of spaces based on a scene flow, updates a probability vector map for clustered spaces, identifies a stationary background based on the updated probability vector map, and estimates ego motion information based on the identified stationary background.