Obstacle Boundary Detection via Edge Persistence Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining obstacle boundaries in images are brittle and fail to detect horizontal occlusion boundaries parallel to target motion, and they do not effectively handle partially occluded targets or occlusions that divide the floor into disconnected segments, limiting their applicability in complex scenes.
Innovation Solution
A method that detects objects in images, determines regions where objects are present, identifies edge locations, and finds a common location among these edges to define the boundary of the obstacle, which can be partially composed of these edges, allowing for robust detection of obstacle boundaries without requiring static occlusion boundaries to coincide with intensity edges.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If existing methods assume static occlusion boundaries coincide with intensity edges, then detection simplicity is maintained, but detection accuracy deteriorates in complex scenes with horizontal occlusions
Solution Approach 1:
The method segments the occlusion boundary detection problem into two independent components: (1) detecting persistent edges from moving targets, and (2) detecting common locations across multiple targets. This segmentation allows the system to handle horizontal occlusions without requiring intensity edge coincidence, resolving the contradiction between simplicity and accuracy.
Solution Approach 2:
The invention transitions from relying solely on intensity edge information (2D image space) to incorporating temporal persistence across multiple frames (adding time dimension). By finding common locations of persistent edges across multiple targets and frames, the system achieves accurate horizontal occlusion detection without being constrained by intensity edge alignment.
2Device complexity
If methods rely on initial segmentation and target path assumptions, then computational complexity is reduced, but reliability deteriorates due to brittleness
Solution Approach 1:
The method employs feedback through temporal persistence analysis, where edges are tracked across multiple frames and their persistence is evaluated. This feedback mechanism allows the system to reliably identify occlusion boundaries without depending on initial segmentation quality or target path assumptions, resolving the contradiction between complexity and reliability.
Solution Approach 2:
The system uses the moving targets themselves to provide the information needed for occlusion detection. By analyzing persistent edges from multiple targets that naturally move through the scene, the method eliminates the need for external segmentation inputs or path planning, achieving both simplicity and robustness.
3Measurement precision
If methods detect all edges in regions, then completeness of boundary detection is improved, but false positive rate increases due to over-segmentation
Solution Approach 1:
The method performs preliminary filtering by identifying persistent edges across multiple frames before final boundary determination. This preliminary action eliminates non-persistent edges early in the process, ensuring that only relevant edges contribute to the occlusion boundary, thus maintaining completeness while reducing false positives.
Solution Approach 2:
The invention merges edge information from multiple targets by finding common locations. By combining persistent edges from multiple independent targets, the system achieves complete boundary detection while the requirement for commonality across targets naturally filters out false positives, resolving the contradiction between completeness and false positive rate.
Data Source
AI summary
A method of determining a boundary of an obstacle which occludes an object in an image, is disclosed. At least one object in the image and a further image captured by a camera is determined. A plurality of regions in which the at least one object is detected, is determined from the images. Locations of edges of the plurality of regions are determined A common location is determined from the determined edges. The boundary of the obstacle is determined, where the boundary comprises a part of the edges at the common location.