Object Recognition via Voting Maps for Small Distant Objects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection methods in driver assistance systems struggle with robustly detecting small and distant objects due to low resolution and missing edge information, especially in varying environments and lighting conditions, which limits their effectiveness in camera-based systems.
Innovation Solution
A device and method for object recognition that subdivides input images into zones and generates saliency-guided voting maps to identify candidate regions without prior knowledge of the environment, allowing for the detection of small and distant objects by emphasizing visually distinct areas and combining binarized voting maps to refine hypotheses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional object detection methods are used, then detection of high-resolution objects is achieved, but detection of small and distant objects fails due to low resolution and missing edge information
Solution Approach 1:
The patent divides the image into multiple zones and further segments each zone into patches. This segmentation allows the system to process different regions with appropriate detail levels, preserving edge information in critical areas while maintaining overall detection capability for distant objects.
Solution Approach 2:
The patent introduces voting maps as an additional dimensional representation. Instead of relying solely on pixel intensity and edge information in the original image space, the system creates a voting map space where each patch contributes votes to potential object locations, effectively adding a new dimension for object hypothesis generation that compensates for lost edge information.
2Area of stationary object
If camera-based systems monitor entire area in front of vehicle, then coverage is improved, but detection of small distant objects deteriorates due to resolution limitations
Solution Approach 1:
By dividing the large monitoring area into multiple zones and patches, the system can apply different processing strategies to different regions. Distant regions are processed with zone-based voting maps that aggregate information across larger areas, while closer regions receive more detailed patch-level analysis, maintaining precision across the entire field of view.
Solution Approach 2:
The patent applies voting map generation selectively to specific zones and patches rather than uniformly across the entire image. This partial action approach focuses computational resources on regions where small distant objects are most likely to appear, improving detection precision without requiring excessive processing of the entire monitoring area.
3Reliability
If existing object detection methods are used, then training data can be utilized, but robust detection of small and distant objects fails without prior environmental knowledge
Solution Approach 1:
The voting map generation process is self-adapting to different environmental conditions. Each patch automatically generates votes based on its local characteristics and the statistical properties of its zone, without requiring external training data or prior environmental knowledge. The system serves itself by learning patterns directly from the input image through the voting aggregation process.
Solution Approach 2:
The zone-based voting map approach serves multiple functions simultaneously: it detects objects at various distances, adapts to different lighting and weather conditions, and works across diverse environments without retraining. The same voting map mechanism universally handles both near and distant objects, making the system versatile across different operational conditions.
Data Source
AI summary
A device for object recognition of an input image includes: a patch selector configured to subdivide the input image into a plurality of zones and to define a plurality of patches for the zones; a voting maps generator configured to generate a set of voting maps for each zone and for each patch, and to binarize the generated set of voting maps; a voting maps combinator configured to combine the binarized set of voting maps; and a supposition generator configured to generate and refine a supposition out of or from the combined, binarized set of voting maps.