Temporal Mask Filtering for Multi-Target Image Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition systems struggle with identity switch errors, particularly when multiple targets are in proximity, leading to incorrect assignment of identifiers due to turbid feature quantities caused by occlusions.
Innovation Solution
An image recognition apparatus uses a two-stage method involving detection and feature extraction, predicting masks based on temporal information to filter feature quantities within bounding boxes, thereby reducing identity transfer errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If feature quantities are extracted from detection objects with multiple targets in proximity, then detection coverage is improved, but feature quantity clarity deteriorates due to occlusions causing identity switch errors
Solution Approach 1:
The patent segments the feature extraction process by dividing the detection object into multiple regions based on predicted masks from previous time steps. Each region is processed separately to extract feature quantities, preventing contamination from occluded areas and maintaining feature clarity even when multiple targets are in proximity.
Solution Approach 2:
The patent performs preliminary mask prediction using temporal information before extracting feature quantities. By predicting which regions should contain targets based on previous frames, the system can pre-identify and isolate relevant features, preventing identity switch errors before they occur during feature extraction.
2Reliability
If masks are predicted based on temporal information, then identity tracking accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent uses periodic mask prediction based on temporal information from previous time steps. By leveraging temporal patterns and predicting masks at regular intervals rather than continuously, the system maintains identity tracking accuracy while reducing computational burden compared to continuous prediction approaches.
Data Source
AI summary
An image recognition apparatus that recognizes a target with respect to image data by detecting a plurality of targets with respect to image data and outputting a plurality of detection objects that is based on the detected plurality of targets, extracting respective feature quantities from the output plurality of detection objects, outputting, with respect to each of the detection objects, a filtered feature, which is a feature quantity obtained by filtering the feature quantity extracted from each of the detection objects, based on a first mask for current time for each detection object predicted at previous time, and predicting the first mask for next time for each of the detection objects.