Manifold Learning for Autonomous Vehicle Object Pose Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer vision systems for autonomous vehicles struggle to identify unknown objects and objects in unfamiliar poses, leading to potential malfunction and ineffective route planning.
Innovation Solution
A method and system that utilize manifold learning to map images to a 3D manifold space, establishing separate object clusters for different categories and poses, allowing for the estimation of unknown object categories and poses based on their proximity to trained clusters, and enabling improved route planning and vehicle operation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a CNN is trained to identify known objects in specific poses, then object detection accuracy for trained categories is improved, but the system fails to identify unknown objects or objects in untrained poses
Solution Approach 1:
The patent transforms object recognition from traditional 2D image space to a 3D manifold space where objects are represented by multiple views. This dimensional transformation allows the system to capture objects from various angles and poses, enabling recognition of unknown poses by mapping them to the learned 3D manifold structure rather than relying on fixed trained poses.
Solution Approach 2:
The system creates a universal object representation framework that handles both known and unknown objects, poses, and categories simultaneously. The 3D manifold space serves as a universal coordinate system where any object view can be mapped and compared, making the system versatile across different object types and orientations without requiring separate trained models for each.
2Quantity of substance
If the CNN is trained on limited object categories and poses, then training data requirements are reduced, but the system cannot generalize to unseen object variations
Solution Approach 1:
The system performs preliminary construction of a 3D manifold space during training, creating a comprehensive geometric framework that anticipates and accommodates future unknown object views. By pre-establishing this flexible 3D representation structure with multiple object views, the system prepares a robust foundation that can generalize to unseen objects and poses without requiring extensive additional training data.
3Device complexity
If traditional object detection methods are used, then implementation simplicity is maintained, but the system cannot provide accurate 3D structure understanding for autonomous navigation
Solution Approach 1:
The patent segments the object recognition process into distinct components: extracting multiple views of objects, mapping these views to a 3D manifold space, and performing classification in this transformed space. This segmentation allows the complex 3D understanding task to be broken down into manageable steps while preserving comprehensive spatial information that traditional single-view methods lose.
Data Source
AI summary
An autonomous vehicle vision system for estimating a category of a detected object in an object pose unknown to the system includes a neural network to apply a mapping process to a region of interest in an image including the detected object in the object pose to obtain a point in a 3D manifold space. The system includes an object detector to estimate the category of the detected object in the object pose in the region of interest based on a relationship between the point representing the detected object in the object pose and a plurality of separate object clusters in the 3D manifold space. The system further includes a planner to select an improved route based on a predicted behavior of the category of the detected object in the object pose. The system also includes a controller to control operation of an autonomous vehicle according to the improved route.