Object Re-Identification Using Selective Neural Network Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object re-identification technologies, particularly using neural networks, face challenges in identifying objects when they are partially occluded or viewed from different angles, leading to difficulties in tracking objects across scenes monitored by different cameras.
Innovation Solution
The method involves training multiple neural networks on different sets of anatomical features represented by reference vectors, selecting the most suitable network based on the input vector's similarity to these reference vectors, and ensuring the input data meets quality conditions to improve re-identification accuracy without requiring complex algorithms for estimating non-depicted object parts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single neural network is trained on complete body structures, then re-identification accuracy is improved for fully visible objects, but performance deteriorates when objects are partially occluded or show different anatomical features
Solution Approach 1:
The patent divides the re-identification system into multiple specialized neural networks, each trained on specific anatomical features (e.g., full body, upper body, lower body). This segmentation allows the system to handle different visibility conditions with dedicated models rather than one general model, directly resolving the contradiction between accuracy for complete objects and adaptability to partial objects.
Solution Approach 2:
The system changes the training parameters (anatomical feature sets) of neural networks to create specialized models for different object visibility conditions. By varying the training data parameters to include different anatomical feature combinations, the system achieves both high accuracy for specific conditions and broad adaptability across multiple conditions.
2Adaptability or versatility
If multiple neural networks are trained on different anatomical features, then adaptability to different object visibility conditions is improved, but system complexity increases
Solution Approach 1:
The system performs preliminary classification of input images to determine which anatomical features are visible, then selects the appropriate pre-trained neural network accordingly. This preliminary action avoids the need to run multiple networks simultaneously, reducing computational complexity while maintaining adaptability to different visibility conditions.
Solution Approach 2:
The patent introduces an intermediary component (selection mechanism) that mediates between the multiple neural networks and the input data. This intermediary evaluates the input image characteristics and routes it to the most suitable network, managing the complexity of having multiple networks without requiring complex integration logic.
3Measurement precision
If neural networks are trained on homogeneous anatomical features, then re-identification accuracy for specific feature sets is improved, but performance on diverse object presentations deteriorates
Solution Approach 1:
The patent segments the training data into homogeneous anatomical feature sets (full body, upper body, lower body, etc.) and trains separate neural networks on each segment. This allows each network to achieve high accuracy for its specific feature set while the collection of networks provides versatility across diverse object presentations.
Solution Approach 2:
The system achieves universality by creating a multi-functional ensemble of neural networks, where each network specializes in a particular anatomical feature configuration. The collective system can handle diverse object presentations by selecting the appropriate specialized network, combining specialization with universal applicability.
Data Source
AI summary
A method of object re-identification in images of objects comprises providing a plurality of neural networks for object re-identification, wherein each of the plurality of neural networks is trained on image data with different sets of anatomical features, each set being represented by a reference vector; receiving a plurality of images of objects and an input vector representing anatomical features that are depicted in all of the plurality of images; comparing the input vector with the reference vectors for determining, according to a predefined condition, the most similar reference vector; and inputting image data of the plurality of objects to the neural network represented by the most similar reference vector for determining whether the plurality of objects have the same identity.