Pedestrian Re-Identification via Temporal Clustering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing pedestrian re-identification systems are prone to failure due to external noise interference from factors like posture, illumination, and shooting angle, which affect the accuracy of feature distances and lead to incorrect identification.

Innovation Solution

A method that detects pedestrians in video frames, extracts historical pedestrians within a preset time period, calculates feature distances, clusters IDs based on ascending order, and assigns IDs based on maximum counts to improve re-identification accuracy by suppressing noise interference.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If pedestrian re-identification is based on appearance characteristics, then identification can be performed using image data, but external noise interference from posture, illumination, and shooting angle greatly affects accuracy and causes identification failure

Engineering Contradiction:
Improvepedestrian re-identification accuracyVSAvoidexternal noise interference
Core Design Contradiction:
ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent segments the pedestrian identification process into multiple independent feature dimensions (appearance features, gait features, trajectory features) rather than relying on a single appearance-based identification. This segmentation allows the system to evaluate multiple aspects of pedestrian behavior and appearance separately, reducing the impact of noise in any single dimension on overall identification accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent creates a universal re-identification framework that can handle multiple types of features (appearance, gait, trajectory) and multiple noise sources through a unified feature fusion mechanism. This multi-functional approach allows the same system to process different feature types and suppress various external noise interferences simultaneously, improving reliability across diverse scenarios.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Productivity

If feature distance calculation is used for pedestrian matching, then identification speed can be maintained, but noise interference affects the accuracy of feature distances and leads to incorrect identification

Engineering Contradiction:
Improveidentification speedVSAvoidfeature distance accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent performs preliminary feature extraction and normalization before distance calculation, preparing robust feature representations that are less sensitive to noise. By pre-processing features to reduce noise impact before the actual distance computation, the system maintains fast identification speed while improving the accuracy of feature distance measurements for pedestrian matching.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent transforms the feature space by applying normalization and weighting adjustments to change the parameters of feature representation. This parameter transformation optimizes the feature distance calculation by emphasizing reliable features and minimizing the influence of noisy features, thereby improving measurement precision without sacrificing identification speed.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3761221B1Pedestrian re-identification method and apparatus
Publication Date: 2023.12.13 BEIJING JINGDONG SHANGKE INFORMATION TECH CO LTD
  • EP3761221B1 patent drawingFigure 1
  • EP3761221B1 patent drawingFigure 2
  • EP3761221B1 patent drawingFigure 3

AI summary

The present disclosure provides a pedestrian re-identification method and apparatus. The pedestrian re-identification apparatus detects a pedestrian to be identified in a specified video frame, and within a specified time period before the capture time of the specified video frame, extracts all pedestrians appearing in the captured video frame and taking them to be historical pedestrians; calculating the feature distances between the pedestrian to be identified and the historical pedestrians; in order of feature distance, from small to large, extracting historical pedestrian identifications corresponding to a predetermined number of feature distances, and clustering the extracted historical pedestrian identifications so as to re-identify the pedestrians to be identified. The present disclosure effectively suppresses external noise interference and improves the success rate of pedestrian re-identification.