Pedestrian Re-Identification via Local Feature Attention

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Pedestrian re-identification technologies face challenges in identifying pedestrians across different cameras due to variations in posture, appearance, and shooting distance, often lacking usable face images, and existing methods are inefficient in capturing and combining local feature sequences effectively.

Innovation Solution

A pedestrian re-identification method and apparatus based on local feature attention, utilizing a multi-head attention mechanism neural network to extract video image feature sequences, combining local pedestrian feature sequences through fully connected layers and activation functions, and iteratively training a pedestrian re-identification neural network with residual connections to predict position coordinates and perform identification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional pedestrian re-identification methods are used, then the system can identify pedestrians across different cameras, but the identification accuracy is low due to variations in posture, appearance, and shooting distance

Engineering Contradiction:
Improvepedestrian identification accuracyVSAvoidadaptability to posture, appearance, and shooting distance variations
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent divides the pedestrian image into multiple local regions (head, body, legs, etc.) and extracts features from each region separately using region proposal networks. This segmentation allows the system to focus on discriminative local features that remain consistent across different postures and appearances, thereby improving identification accuracy while adapting to variations in shooting conditions

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different processing and attention mechanisms to different local regions of the pedestrian image. By using region-specific feature extraction and attention weighting, the system emphasizes important local features (such as head shape, body proportions) that are more stable across variations, thus improving both accuracy and adaptability to different shooting distances and postures

Inventive Principle:
Principle #3Local quality

2Measurement precision

If face images are used for pedestrian identification, then high accuracy can be achieved, but usable face images cannot be acquired in most surveillance cases

Engineering Contradiction:
Improvepedestrian identification accuracyVSAvoidavailability of usable face images
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The patent extracts and removes the dependency on face images by focusing exclusively on full-body pedestrian features. Through region proposal networks and attention mechanisms, the system extracts discriminative features from the entire pedestrian body (head, torso, legs, footwear), achieving accurate identification without requiring face visibility, thus making the system operational in typical surveillance scenarios where faces are not captured

Inventive Principle:
Principle #2Taking out (Extraction)

3Measurement precision

If multi-head attention mechanism is used to capture local feature sequences, then the system can effectively combine local features, but the computational complexity increases

Engineering Contradiction:
Improvelocal feature sequence combination accuracyVSAvoidneural network computational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary region proposal and local feature extraction before applying the multi-head attention mechanism. By pre-segmenting the image into relevant regions and extracting key features in advance, the system reduces the input complexity for the attention mechanism, allowing it to focus on combining features effectively without being overwhelmed by raw pixel data, thus balancing accuracy and computational complexity

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS11699290B1Pedestrian re-identification method and apparatus based on local feature attention
Publication Date: 2023.07.11 SHANGHAI ENFLAME TECH CO LTD
  • US11699290B1 patent drawing
  • US11699290B1 patent drawing
  • US11699290B1 patent drawing

AI summary

Disclosed are a pedestrian re-identification method and apparatus based on local feature attention. The method includes the following steps: S1: obtaining an original surveillance video image data set, and dividing the original surveillance video image data set into a training set and a test set in proportion; and S2: performing image enhancement on the original surveillance video image training set to obtain enhanced images, and converting the enhanced images into sequence data. The pedestrian re-identification technology based on local feature attention uses a multi-head attention mechanism neural network to capture, extract video image feature sequences and replace convolution kernels in a convolutional neural network, uses fully connected layers and an activation function to combine local pedestrian feature sequences into complete pedestrian feature sequences through a weight matrix, performs prediction on the obtained pedestrian feature sequences, outputs position coordinates of pedestrians in the images and selects pedestrians to realize pedestrian re-identification.