Particle Filter Target Tracking with Color Histogram Correction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Shape-based object tracking methods, such as the Condensation algorithm, face challenges in distinguishing target objects from others with similar shapes, leading to errors in likelihood observation and reduced tracking accuracy, especially when multiple objects with similar shapes are present in the scene.

Innovation Solution

An object tracker that generates particles based on an estimated existence probability distribution in a particle filter, using a sampling unit to define candidate contours in a parameter space, an observation unit to apply a motion model and observe likelihood, and a correction unit to adjust likelihood based on color information comparison between image frames, thereby improving the accuracy of existence probability distribution estimation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If shape-based tracking is used to track target objects, then tracking can be performed using simple shape matching, but the system cannot distinguish between target objects and other objects with similar shapes, leading to tracking errors

Engineering Contradiction:
Improvetracking implementation simplicityVSAvoidtracking accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The invention extends the tracking parameters from shape-only to include both shape and color information. The particle representation is enhanced to include color attributes, and the likelihood function is modified to incorporate color distance calculations. This allows the system to distinguish between objects with similar shapes by comparing their color characteristics, thereby resolving the contradiction between simple shape-based tracking and accurate object discrimination.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If color information is incorporated to distinguish similar shapes, then tracking accuracy improves, but computational complexity increases

Engineering Contradiction:
Improvetracking accuracyVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The invention applies partial action by selectively incorporating color information only for particles that are shape-matched, rather than computing color distances for all particles. The correction step is applied conditionally based on shape similarity threshold, which reduces unnecessary computations while still achieving the goal of distinguishing similar-shaped objects through color information.

Inventive Principle:
Principle #16Partial or excessive action

3Adaptability or versatility

If multiple objects with similar shapes are present in the scene, then the tracking scenario becomes more complex, but shape-based methods alone cannot differentiate between them

Engineering Contradiction:
Improvehandling multiple objects capabilityVSAvoidobject identity distinction
Core Design Contradiction:
Adaptability or versatilityVSLoss of information

Solution Approach 1:

The invention enhances the particle representation to include color attributes in addition to shape parameters. The likelihood function is extended to incorporate color distance calculations, allowing the system to maintain distinct identities for multiple objects with similar shapes by comparing their color characteristics. This prevents object identity loss in cluttered scenes with multiple similar-shaped objects.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP2192549B1Target tracking device and target tracking method
Publication Date: 2013.07.17 SONY INTERACTIVE ENTERTAINMENT LLC
  • EP2192549B1 patent drawingFigure 1~2
  • EP2192549B1 patent drawingFigure 3
  • EP2192549B1 patent drawingFigure 4

AI summary

Referring to Fig. 5, the tracking start-and-stop determining unit 28 of the tracking processing unit 26 determines the starting and stopping of tracking based on predetermined conditions. During the tracking, the sampling unit 29 creates or eliminates particles using the probability density distribution estimated for an immediately preceding image frame. The observation unit causes the particles to make a transition according to a predetermined motion model and observes the likelihoods of a candidate curves defined by the particles. The correction unit 32 compares the color histogram of an area located at a predetermined position relative to the candidate curve with the color histogram of an area similarly located relative to the curve determined as a result of tracking in the immediately preceding image frame. If the matching score is equal to or higher than a predetermined value, the likelihood of the candidate curve is increased by correction. The result obtaining unit 34 estimates the probability density distribution based on the likelihood thus corrected.