Human Identification Model Training via Selective Camera Subset
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-camera systems in service robots face challenges in efficiently processing increasing amounts of data for human recognition and re-identification, leading to computational burdens and suboptimal performance.
Innovation Solution
A device and method for training a model for human identification, which includes a primary training module, a target subset generation module, a feature vector extraction module, a feature vector clustering module, a labeling module, and a secondary training module, to improve human recognition and re-identification performance by selectively processing data from cameras with smaller differences from a source dataset.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a multi-camera system is used to improve human recognition ability, then recognition accuracy and three-dimensional understanding are improved, but the amount of data to be processed and computational complexity increase
Solution Approach 1:
The patent segments the training process into two distinct phases: pre-training on a source dataset and fine-tuning on a target dataset. This segmentation allows the model to first learn general features from diverse data and then specialize for specific camera configurations, reducing the computational burden during deployment while maintaining high recognition accuracy across multiple cameras.
Solution Approach 2:
The patent performs preliminary pre-training of the model on a source dataset before deploying it to the multi-camera system. This preliminary action prepares the model to handle various camera perspectives and conditions, reducing the computational complexity required during actual operation by having the heavy lifting done during the pre-training phase.
2Measurement precision
If data from all cameras is processed to improve re-identification performance, then recognition accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The patent applies partial action by selectively processing data from cameras that provide the most valuable information for re-identification. Rather than processing all camera data equally, the system identifies and processes only the necessary subset, reducing processing time while maintaining or improving re-identification accuracy through targeted data selection.
Solution Approach 2:
The patent applies local quality by treating data from different cameras differently based on their specific characteristics and contributions to re-identification. Each camera's data is processed with appropriate weighting and selection criteria tailored to its local properties, optimizing the balance between accuracy and processing efficiency.
3Measurement precision
If the model is trained with all available camera data, then human identification performance improves, but training time and computational resources increase
Solution Approach 1:
The patent segments the training process into pre-training and fine-tuning phases, with each phase using appropriately selected datasets. This segmentation improves training efficiency by avoiding the need to process all camera data at full detail throughout the entire training process, while still achieving high human identification performance through the combined effect of both phases.
Solution Approach 2:
The patent performs preliminary pre-training on a source dataset before fine-tuning on target data from specific cameras. This preliminary action reduces the overall training time and computational resources required by preparing the model in advance, allowing faster convergence during the final fine-tuning stage while maintaining high identification performance.
Data Source
AI summary
A device, and method thereof, for training a model for human identification are provided may include a primary training module configured to primarily train the model with respect to a pre-prepared source dataset, a target subset generation module configured to generate a target subset by selecting some cameras from among a plurality of cameras included on a service robot, a feature vector extraction module configured to extract feature vectors of the target subset by using the model, a labeling module configured to perform labeling on the feature vectors, and a secondary training module configured to secondarily train the model with respect to a target dataset by using results of the labeling.