Single-Image Traversable Space Mapping for Occluded Surface Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting traversable space in a scene from a single image are limited, as they often rely on visible surfaces and fail to detect occluded areas, requiring resource-intensive computations and relying on synthetic or small-scale training environments, which are not practical for real-world applications.
Innovation Solution
A machine-learning model that predicts the geometry of both visible and occluded traversable surfaces using stereo video sequences, camera poses, and semantic segmentation, allowing for efficient determination of traversable space from a single RGB image, accounting for object movement and footprints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If existing methods predict traversable space using line-of-sight algorithms, then visible surfaces can be detected, but occluded traversable surfaces cannot be detected
Solution Approach 1:
The model performs preliminary depth prediction and surface geometry estimation before path planning, using the predicted traversable space map to guide subsequent navigation decisions. This allows the system to anticipate occluded areas based on visible cues and object footprints, rather than reacting to them during path execution.
Solution Approach 2:
The patent introduces an intermediate representation of traversable space that bridges visible image data and hidden ground surfaces. This intermediate map incorporates object footprint predictions and depth estimates to infer occluded traversable areas, serving as a mediator between limited camera observations and comprehensive environmental understanding.
2Loss of information
If voxel-based or mesh-based methods are used to predict beyond line-of-sight, then occluded traversable space can be estimated, but computational cost increases significantly
Solution Approach 1:
The patent extracts only the essential traversable space information needed for navigation, rather than reconstructing complete 3D voxel or mesh representations. By focusing specifically on ground surface geometry and object footprints, the system achieves occluded area detection with reduced computational overhead.
Solution Approach 2:
The model uses lightweight, disposable predictions of traversable space that are generated on-demand for each image frame rather than maintaining persistent complex spatial data structures. These lightweight predictions are sufficient for navigation purposes and can be rapidly recomputed as new images are captured.
3Reliability
If complete, static training environments are used, then training data can be obtained, but the environment scale is limited or synthetic
Solution Approach 1:
The training system transitions from static environment models to dynamic video sequences, capturing moving objects and changing scenes. This allows the model to learn from temporal variations and real-world dynamics, improving adaptability to diverse and changing environments while maintaining reliable supervision signals.
Solution Approach 2:
The patent develops a universal training framework that works across diverse real-world scenes and conditions, rather than requiring environment-specific training data. The model learns generalizable patterns of traversable space prediction that transfer to various settings, making the system versatile across different real-world applications.
Data Source
AI summary
A model predicts the geometry of both visible and occluded traversable surfaces from input images. The model may be trained from stereo video sequences, using camera poses, per-frame depth, and semantic segmentation to form training data, which is used to supervise an image to image network. In various embodiments, the model is applied to a single RGB image depicting a scene to produce information describing traversable space of the scene that includes occluded traversable. The information describing traversable space can include a segmentation mask of traversable space (both visible and occluded) and non-traversable space and a depth map indicating an estimated depth to traversable surfaces corresponding to each pixel determined to correspond to traversable space.