Object Tracking Device Detection Box Correction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing object tracking technologies fail to accurately correct the size and position of detection boxes, leading to inaccurate trail generation due to limitations in scene variations and background interference.

Innovation Solution

An object tracking apparatus that includes an object detecting unit, a detection box reliability calculating unit, and a detection box position correcting unit, which calculates reliability based on flow information and corrects detection box information to generate a highly accurate trail.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If machine learning dictionary information is used for detection, then various scenes can be learned, but variations of scenes are limited and detection accuracy decreases in varying imaging environments

Engineering Contradiction:

Improvescene variation coverageVSAvoiddetection box accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system performs preliminary object recognition at predetermined frame intervals to identify high-reliability detection boxes in advance. These pre-identified reliable detection boxes are then used as reference for correcting low-reliability detection boxes in subsequent frames, improving overall detection accuracy without requiring extensive re-learning for each scene variation.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system calculates reliability of detection boxes based on flow information and uses this reliability metric to feedback into the correction process. High-reliability detection boxes are identified and their information is used to correct low-reliability boxes, creating a feedback loop that continuously improves detection accuracy across varying scenes.

Inventive Principle:

Principle #23Feedback

2Reliability

If object tracking and object recognition are performed in parallel, then ID information can be corrected, but size and positional information of the rectangle cannot be corrected

Engineering Contradiction:

ImproveID information accuracyVSAvoiddetection box position and size

Core Design Contradiction:

ReliabilityVSMeasurement precision

Solution Approach 1:

The system segments the detection box correction process into distinct components: ID information correction (from parallel tracking/recognition) and position/size correction (from reliability-based correction). By separating these functions, the system can apply different correction strategies for each aspect, enabling comprehensive correction of all detection box parameters.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The reliability calculation based on flow information acts as an intermediary mechanism that bridges ID correction and position/size correction. The reliability metric mediates the selection of high-reliability detection boxes whose position and size information is then used to correct low-reliability boxes, enabling comprehensive correction beyond just ID information.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Device complexity

If detection box information is used directly from machine learning, then processing is simple, but accurate rectangular position cannot be estimated when background is included

Engineering Contradiction:

Improveprocessing complexityVSAvoidrectangular position accuracy

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The system replaces direct machine learning detection with a reliability-based correction mechanism. Instead of relying solely on the machine learning output, flow information (optical flow or depth flow) is used to calculate reliability and correct detection box positions and sizes, substituting the direct detection mechanism with a correction mechanism that achieves higher accuracy.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system changes the parameter used for detection from direct machine learning confidence scores to reliability scores calculated from flow information. This parameter change enables the system to distinguish between high-reliability and low-reliability detection boxes, allowing selective correction of position and size parameters for improved accuracy.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20240153106A1Object tracking device

Publication Date: 2024.05.09 ASTEMO LTD

AI summary

An object of the present invention is to generate a highly accurate trail by correcting detection box information of an object in an object tracking apparatus that generates a trail of an object within a measurement range. According to the present invention, there is provided an object tracking apparatus (100) that generates a trail of an object within a measurement range of a camera (2), the object tracking apparatus (100) including: an object detecting unit (4) that detects an object for each of a plurality of frames acquired by a sensor; a detection box reliability calculating unit (8) that calculates a reliability of a detection box based on flow information between frames of the detection box in which the object has been detected; a detection box position correcting unit (9) that corrects detection box information of a low-reliability detection box, using detection box information of a high-reliability detection box; and a trail generating unit (10) that generates a trail of an object, using the corrected detection box information.