AR Pose Tracking With Motion-Based Key Point Matching

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing pose recognition algorithms fail to connect poses of a particular person across frames in a video sequence, leading to reduced accuracy in multi-person interaction scenarios.

Innovation Solution

A method using a neural network model to track key point sets across frames based on motion trends, determining distances between key points to identify matching sets and generate augmented reality images, ensuring accurate recognition of the same object across frames.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If existing pose recognition algorithms are used, then poses of all persons in each frame can be recognized, but poses of a particular person in a video sequence cannot be connected in series

Engineering Contradiction:
Improverecognition accuracyVSAvoidobject identity continuity
Core Design Contradiction:
Measurement precisionVSLoss of information

Solution Approach 1:

The patent uses feedback by comparing the detected key point sets in the current frame with the predicted key point set (based on motion trends) to determine whether they belong to the same object. The system continuously refines object tracking by feeding back the matching results to maintain identity consistency across frames.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent introduces an intermediary mechanism - the key point set matching based on motion trends - that bridges the gap between frame-by-frame detection and continuous object tracking. This intermediary process enables the connection of pose information across frames by using motion trends as a mediator to link detected key points to their corresponding objects in the video sequence.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If key point sets from multiple objects are incorrectly matched, then recognition accuracy decreases

Engineering Contradiction:
Improveobject identification accuracyVSAvoidtracking algorithm complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces complex mechanical tracking systems with a more elegant solution based on motion trends and distance calculation. Instead of using elaborate tracking mechanisms, the system substitutes a simpler approach that predicts key point positions based on motion trends and identifies objects through distance-based matching, thereby reducing algorithmic complexity while maintaining reliability.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Measurement precision

If distance threshold is set too low, then matching precision increases but more key point sets are missed

Engineering Contradiction:
Improvekey point matching precisionVSAvoidobject recognition speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent applies partial action by using a distance threshold that balances precision and coverage. Rather than setting the threshold to achieve maximum precision (which would miss many valid matches) or maximum coverage (which would include false matches), the system selects a threshold that provides sufficient precision for reliable matching while maintaining adequate recognition speed for practical applications.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3617995B1Augmented reality processing method, object recognition method, and related apparatus
Publication Date: 2026.03.25 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP3617995B1 patent drawingFigure 1~2
  • EP3617995B1 patent drawingFigure 3
  • EP3617995B1 patent drawingFigure 4

AI summary

The present disclosure discloses an object recognition method, including: obtaining a key point set of a first object in a first image; obtaining, through a neural network model, first pose key point sets respectively corresponding to a plurality of objects in the second image; determining a second pose key point set of the first object in the second image according to the key point set and a motion trend of the first object; determining, for any target first pose key point set in the plurality of first pose key point sets, a target distance between the target first pose key point set and the second pose key point set according to at least one first pose key point in the target first pose key point set and at least one second pose key point in the second pose key point set; and using the target first pose key point set as a key point set of the first object in the second image in a case that the target distance satisfies a preset condition. In a multi-person interaction scene, key point sets of a same object can be recognized in a video stream, so that accuracy of recognition is improved.