Video Surveillance Object Tracking Using Correlation Filter Templates

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing video surveillance systems face challenges in accurately tracking faces across different video sequences due to variations in face shape and motion, especially when faces are blocked or blurred.

Innovation Solution

A system and method for tracking objects in video surveillance using a correlation filter to identify the position of an object of interest in a current frame based on template frames from previous frames, even when the object is blocked or blurred.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a single template frame is used for tracking, then the device complexity is reduced, but the tracking accuracy deteriorates due to uncertain motion state and shape variations of faces

Engineering Contradiction:
Improvetracking accuracyVSAvoidtemplate processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the template selection into multiple discrete template frames (at least two different template frames) instead of using a single template. Each template frame captures different features of the object of interest, allowing the system to handle shape variations and motion uncertainties by selecting from multiple templates rather than relying on one complex adaptive template.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary selection of multiple template frames from historical video data before the actual tracking occurs. These templates are pre-processed and stored, so when tracking is needed, the system can directly use these prepared templates without complex real-time adaptation, thus improving accuracy while controlling complexity.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If multiple template frames are used to improve tracking robustness, then the tracking accuracy improves, but the computational complexity increases

Engineering Contradiction:
Improvetracking robustnessVSAvoidcomputational complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent uses a limited number of template frames (at least two, but not an excessive number) to achieve the necessary tracking robustness. This partial action approach avoids the computational burden of using too many templates while still providing enough variation to handle different face shapes and motion states effectively.

Inventive Principle:
Principle #16Partial or excessive action

3Measurement precision

If template features are selected to match specific face shapes, then the tracking accuracy improves for that shape, but the adaptability to different face shapes deteriorates

Engineering Contradiction:
Improvetracking accuracyVSAvoidface shape adaptability
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal tracking solution by using multiple template frames that collectively represent different face shapes and features. Instead of creating specialized templates for each face shape (which would improve accuracy for specific shapes but reduce versatility), the system uses a set of multi-functional templates that can adapt to various face types, achieving both accuracy and adaptability.

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

Data Source

PatentUS12283106B2Systems and methods for video surveillance
Publication Date: 2025.04.22 ZHEJIANG DAHUA TECH CO LTD
  • US12283106B2 patent drawing
  • US12283106B2 patent drawing
  • US12283106B2 patent drawing

AI summary

Systems and methods for tracking an object in video surveillance are provided. A method may include obtaining a video including a plurality of consecutive frames; obtaining a current frame from the plurality of consecutive frames, wherein an object of interest is identified in at least two previous frames of the current frame; obtaining at least two template frames from the at least two previous frames; and identifying a position related to the object of interest in the current frame based on the at least two template frames using a correlation filter.