Pedestrian Re-Identification via Multi-Layer Feature Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Pedestrian re-identification using convolutional neural networks faces challenges due to lack of identifiable features, especially with varying image angles and resolutions, leading to lower identification accuracy compared to face recognition.
Innovation Solution
The method involves extracting global and local feature information from multiple convolutional layers of a convolutional neural network, merging these features to create a more identifiable classification feature, which improves pedestrian re-identification accuracy by fusing features of multiple layers and scales.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If convolutional neural network-based pedestrian re-identification is used, then automated monitoring efficiency is improved, but identification accuracy deteriorates due to lack of identifiable features
Solution Approach 1:
The patent segments the pedestrian image features into multiple convolutional layers, extracting features at different depths. Each layer captures different levels of abstraction, from basic edges to complex patterns, allowing the system to utilize multi-scale features for improved identification accuracy while maintaining automated processing efficiency
Solution Approach 2:
The patent adds a dimensional aspect by extracting features across multiple convolutional layers (depth dimension) rather than relying on single-layer features. This multi-layer feature extraction creates a hierarchical feature space that provides more discriminative information for pedestrian re-identification, resolving the accuracy-automation tradeoff
2Speed
If single-scale feature extraction is used, then processing speed is improved, but identification accuracy deteriorates due to inability to capture multi-scale characteristics
Solution Approach 1:
The patent segments the feature extraction process across multiple convolutional layers, where each layer operates at different scales. This allows parallel extraction of multi-scale features without sequential processing, maintaining speed while improving accuracy through comprehensive feature capture
Solution Approach 2:
The patent merges feature representations from multiple convolutional layers by concatenating or combining the extracted features. This fusion of multi-scale features from different layers creates a comprehensive feature vector that preserves processing efficiency while significantly enhancing identification accuracy
Data Source
AI summary
A method and apparatus for pedestrian re-identification, an electronic device, and a computer-readable storage medium are provided. The method includes that: a pedestrian image to be detected is acquired; global feature information of the pedestrian image to be detected is extracted through multiple convolutional layers of a convolutional neural network; multiple pieces of intermediate feature information of the pedestrian image to be detected are extracted through the multiple convolutional layers of the convolutional neural network respectively, and the multiple pieces of intermediate feature information are merged as local feature information; and the global feature information and the local feature information are assigned as a classification feature of the pedestrian image to be detected, and a classification result of the pedestrian image to be detected is determined according to the classification feature.