Feature Point Tracking via Inter-Frame Prediction

Solution Overview

Problem

Current feature point tracking methods in video processing are complex and inefficient, particularly in determining camera extrinsic parameters for applications like structure from motion and panorama processing, due to the lack of effective inter-frame prediction techniques.

Innovation Solution

An apparatus and method for feature point tracking using inter-frame prediction, which involves a graphics processor architecture that includes a graphics processing engine capable of predicting feature points between frames, enabling efficient detection and matching of key points across video streams.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional feature point tracking methods are used, then feature points can be detected and matched between image pairs, but the process becomes complex and inefficient, especially in determining camera extrinsic parameters

Engineering Contradiction:

Improvefeature point tracking efficiencyVSAvoidtracking method complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent applies preliminary action by using inter-frame prediction to pre-estimate the positions of feature points in the current frame based on motion information from previous frames. This preliminary estimation provides initial candidates for feature point matching, significantly reducing the search space and computational complexity of traditional tracking methods while improving processing efficiency

Inventive Principle:

Principle #10Preliminary action

2Productivity

If inter-frame prediction is implemented, then feature point tracking efficiency improves, but additional processing steps and computational resources are required

Engineering Contradiction:

Improvefeature point tracking speedVSAvoidprocessing pipeline complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent uses motion vectors as an intermediary element to bridge previous frame information and current frame feature point prediction. By introducing motion compensation technology as a mediator, the system efficiently transfers spatial-temporal correlations between frames without requiring complex direct analysis, thus improving tracking speed while keeping the additional processing steps manageable

Inventive Principle:

Principle #24Intermediary (Mediator)

3Measurement precision

If key-point descriptors are matched using approximate nearest-neighbors method, then feature points can be matched between image pairs, but the accuracy and efficiency of camera extrinsic parameter determination is limited

Engineering Contradiction:

Improvecamera extrinsic parameter accuracyVSAvoidparameter determination time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by using inter-frame prediction to pre-estimate the positions of feature points in the current frame based on motion information from previous frames. This preliminary estimation provides initial candidates for feature point matching, significantly reducing the search space and computational complexity of traditional tracking methods while improving processing efficiency

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11470343B2Apparatus and method for feature point tracking using inter-frame prediction

Publication Date: 2022.10.11 INTEL CORP

AI summary

Apparatus and method for inter-frame prediction. For example, one embodiment of a method comprises extracting a first set of key points from a current frame and a second set of key points from a reference frame; projecting a key point p into the reference frame as p′ based on a motion vector associated a region in the current frame in which key point p is found; locating n key points in the reference frame which are closest to p′; and matching one of the n key points in the reference frame to key point p in the current frame based on descriptor distances between each of the n key points and p′.