Object Recognition Using Binary Mask Dictionary for Occlusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning algorithms for object recognition suffer from degraded performance when objects are partially occluded, as they fail to effectively suppress the influence of occlusion regions on feature elements, leading to inconsistencies in similarity calculations between recognized objects and database features.
Innovation Solution
The proposed solution involves a Pairwise Differential Siamese Network (PDSN) structure that learns a mapping relationship between occlusion regions and feature elements affected by occlusions, using a binary mask dictionary to suppress interference from occluded features, thereby improving recognition accuracy by focusing on non-occluded parts of the object.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current deep learning algorithms are used for object recognition, then recognition speed and simplicity are maintained, but recognition accuracy degrades when objects are partially occluded
Solution Approach 1:
The patent segments the object recognition process into multiple components: occlusion detection module that identifies occluded regions, feature extraction module that processes both occluded and non-occluded features separately, and a recognition module that integrates results. This segmentation allows the system to handle occluded objects by processing different regions with appropriate methods, thereby improving recognition accuracy without requiring complete algorithm redesign
Solution Approach 2:
The patent introduces an intermediary occlusion detection mechanism that acts as a mediator between the input image and the recognition algorithm. This intermediary component detects occlusion regions and provides guidance to the feature extraction and recognition modules, enabling them to adjust their processing accordingly. The intermediary layer resolves the contradiction by adding a specialized component rather than complicating the entire recognition system
2Measurement precision
If occlusion regions are suppressed to improve recognition accuracy, then recognition accuracy for occluded objects improves, but recognition accuracy for non-occluded objects may be affected
Solution Approach 1:
The patent implements a dynamic recognition system that adapts its processing strategy based on occlusion detection results. When occlusion is detected, the system dynamically adjusts by suppressing occluded region features and emphasizing non-occluded region features. When no occlusion is detected, the system uses standard processing. This dynamic adaptation ensures high recognition accuracy for both occluded and non-occluded objects, maintaining reliability across different scenarios
Solution Approach 2:
The patent applies different processing qualities to different regions of the object based on occlusion status. Occluded regions have their feature weights reduced or suppressed, while non-occluded regions maintain full feature contribution. This local quality adjustment ensures that the recognition system focuses on reliable features while minimizing the impact of occluded regions, thereby improving accuracy for occluded objects without compromising overall recognition consistency
Data Source
AI summary
An object recognition method is provided. The method includes: detecting an occlusion region of an object in an image, to obtain a binary image; obtaining occlusion binary image blocks; querying a mapping relationship between occlusion binary image blocks and binary masks included in a binary mask dictionary to obtain binary masks corresponding to the occlusion binary image blocks; synthesizing the binary masks queried based on each of the occlusion binary image blocks, to obtain a binary mask corresponding to the binary image; and determining a matching relationship between the image and a prestored object image, based on the binary mask corresponding to the binary image, a feature of the prestored object image, and a feature of the to-be-recognized image.