Person Tracking State Filtering for Action Detection Accuracy

Solution Overview

Problem

In image analysis, especially when persons overlap, the accuracy of person detection and tracking declines, leading to increased probabilities of false detection and non-detection due to swapped tracking IDs and overlap-induced noise.

Innovation Solution

An image processing apparatus that acquires and tracks persons in images, using state information to determine whether to utilize tracking results based on action detection processing, thereby excluding suspended tracking states in parameter calculation and including them in independent action determination to reduce false detections.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If tracking results with lower detection and tracking accuracy due to overlap between persons are all used, then the amount of information to be used in action detection increases, but the accuracy of information to be used in action detection decreases

Engineering Contradiction:

Improveamount of informationVSAvoidaccuracy of information

Core Design Contradiction:

Quantity of substanceVSMeasurement precision

Solution Approach 1:

The patent applies local quality by differentiating the treatment of tracking results based on their quality characteristics. High-quality tracking results (with high detection and tracking accuracy) are used for parameter calculation, while low-quality tracking results (with lower accuracy due to overlap) are used for independent action determination. This selective application based on local quality characteristics resolves the contradiction by matching the appropriate tracking results to the appropriate detection tasks.

Inventive Principle:

Principle #3Local quality

2Measurement precision

If tracking results with lower detection and tracking accuracy due to overlap between persons are all ignored, then the accuracy of information to be used in action detection improves, but the amount of information to be used in action detection decreases

Engineering Contradiction:

Improveaccuracy of informationVSAvoidamount of information

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent applies local quality by differentiating the treatment of tracking results based on their quality characteristics. High-quality tracking results (with high detection and tracking accuracy) are used for parameter calculation, while low-quality tracking results (with lower accuracy due to overlap) are used for independent action determination. This selective application based on local quality characteristics resolves the contradiction by matching the appropriate tracking results to the appropriate detection tasks.

Inventive Principle:

Principle #3Local quality

3Productivity

If tracking results are used for action detection, then action detection can be performed, but false detection and non-detection increase due to swapped tracking IDs and overlap-induced noise

Engineering Contradiction:

Improveaction detection capabilityVSAvoidfalse detection and non-detection

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent applies local quality by differentiating the treatment of tracking results based on their quality characteristics. High-quality tracking results (with high detection and tracking accuracy) are used for parameter calculation, while low-quality tracking results (with lower accuracy due to overlap) are used for independent action determination. This selective application based on local quality characteristics resolves the contradiction by matching the appropriate tracking results to the appropriate detection tasks.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent introduces an intermediary mechanism (the detection control unit) that mediates between the tracking results and the action detection processes. This intermediary selectively determines which tracking results to use for which detection task, acting as a filter that prevents swapped tracking IDs and overlap-induced noise from causing false detections while still allowing action detection to proceed.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20230222833A1Image processing apparatus, image processing method, and storage medium

Publication Date: 2023.07.13 CANON KK

AI summary

An image processing apparatus includes at least one processor and at least one memory storing executable instructions which, when executed by the at least one processor, cause the image capturing apparatus to acquire an image of at least one person, detect at least one person in the acquired image, control processing for tracking the at least one detected person, and control processing for detecting an action of the at least one detected person based on a result of tracking the at least one detected person. The result of tracking the at least one detected person includes state information representing a tracking state of the at least one detected person. Whether to use the result of tracking the at least one detected person is based on a type of the action detection processing and the state information.