Object Recognition Using Attribute-Based Detection Grouping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional non-maximum suppression methods in object detection rely solely on geometric overlap and detection quality, leading to ambiguities and errors in recognizing multiple objects, particularly in scenarios with overlapping or nearby objects, which can result in incorrect suppression of valid detections.
Innovation Solution
The proposed method enhances non-maximum suppression by incorporating additional semantic attributes such as direction of movement, orientation, distance from a sensor, and interaction between objects, using a temporally or spatially resolved distance measure to group and differentiate raw detections, thereby improving object recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional non-maximum suppression uses only geometric overlap and detection quality, then the calculation complexity remains low, but the object recognition accuracy deteriorates due to ambiguities in complex scenarios
Solution Approach 1:
The patent segments the object recognition problem by introducing multiple independent attributes (geometric overlap, detection quality, semantic attributes like direction of movement, orientation, distance) that can be evaluated separately. Each attribute contributes independently to the final suppression decision, allowing the system to improve accuracy through multiple dimensions without creating a monolithic complex calculation structure.
Solution Approach 2:
The patent transitions from two-dimensional suppression (geometric overlap only) to multi-dimensional suppression by adding semantic attributes as new dimensions. Attributes such as direction of movement, orientation, and distance from sensor create additional evaluation dimensions, enabling more precise differentiation between overlapping objects while maintaining manageable computational complexity through modular attribute processing.
2Reliability
If non-maximum suppression uses a low overlap threshold value, then double detections are reduced, but correct detections are suppressed when multiple objects appear in neighboring regions
Solution Approach 1:
The patent changes the suppression parameters from a single geometric overlap threshold to multiple attribute-based thresholds. By incorporating semantic attributes (direction of movement, orientation, distance), the system can dynamically adjust suppression decisions based on the specific characteristics of each detection, preserving correct detections that would otherwise be lost while still eliminating double detections.
Solution Approach 2:
The patent replaces the mechanical geometric overlap-based suppression mechanism with a more sophisticated attribute-based evaluation system. Instead of relying solely on spatial intersection calculations, the system substitutes in semantic attribute comparisons (direction, orientation, distance) that provide more nuanced discrimination between true multiple objects and false duplicate detections.
Data Source
AI summary
A method for recognizing an object from input data is disclosed. Raw detections are carried out in which at least one attribute in the form of a detection quality is determined for each raw detection. At least one further attribute for each raw detection is determined. A temporally or spatially resolved distance measure is determined for at least one attribute of the raw detections. Raw detections of a defined distance measure are combined to form a group of raw detections. The object is recognized from a group with at least one raw detection with the smallest distance measure of the at least one attribute in comparison with another raw detection, or from a group with at least one raw detection which were combined by combining at least two raw detections with the smallest distance measure of the at least one attribute to form said one raw detection.