Dynamic Pyramid Layer Adjustment for Optical Flow Tracking

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing optical flow estimation methods, such as the Lucas Kanade tracker, are inefficient in terms of computational power and accuracy, especially when dealing with varying object sizes and movements, as they require a fixed number of pyramid layers, which can lead to inaccurate feature tracking and increased computational requirements.

Innovation Solution

A dynamic model adaptation of the pyramidal implementation of optical flow trackers, where the number of pyramid layers is adjusted based on movement speed, camera position, and feature distance, allowing for efficient and accurate feature tracking by increasing or decreasing the number of layers according to the situation, using algorithms like Lucas Kanade and incorporating odometry and camera-based motion detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a fixed number of pyramid layers is used in optical flow tracking, then the computational requirements are predictable and manageable, but the tracking accuracy deteriorates when dealing with varying object sizes and large displacements

Engineering Contradiction:
Improvetracking accuracyVSAvoidcomputational requirements
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies dynamics by making the number of pyramid layers variable rather than fixed. The system dynamically adjusts the number of pyramid layers based on detected motion magnitude and object characteristics, allowing it to increase layers for large displacements to maintain accuracy while reducing layers for small motions to save computational resources.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes the parameter of pyramid layer count adaptively. By monitoring motion estimates and object size during tracking, the system modifies the number of pyramid layers in real-time, transitioning between different computational configurations to optimize both accuracy and efficiency for varying scene conditions.

Inventive Principle:
Principle #35Parameter changes

2Reliability

If the number of pyramid layers is increased to handle large displacements, then the feature tracking accuracy improves, but the computational power requirements increase

Engineering Contradiction:
Improvefeature tracking reliabilityVSAvoidcomputational power
Core Design Contradiction:
ReliabilityVSPower

Solution Approach 1:

The system dynamically adjusts pyramid layer count based on actual tracking needs. When large displacements are detected, it increases layers to maintain reliable feature tracking. When motions are small, it reduces layers to conserve computational power, thus achieving reliability only when necessary.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent applies partial action by using only the necessary number of pyramid layers for the current tracking situation. Instead of always using the maximum number of layers, it employs just enough layers to achieve reliable tracking for the current motion magnitude, avoiding excessive computational expenditure.

Inventive Principle:
Principle #16Partial or excessive action

3Adaptability or versatility

If a fixed pyramidal structure is used for optical flow estimation, then the algorithm is simple to implement, but it cannot adapt to changing motion patterns and object sizes

Engineering Contradiction:
Improveadaptability to motion patternsVSAvoidalgorithm complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent transforms the static pyramidal structure into a dynamic one that adapts to changing motion patterns. The system monitors object size, motion magnitude, and tracking confidence to adjust pyramid layer count, enabling it to handle diverse scenarios from stationary objects to rapid movements while maintaining algorithmic coherence.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3635684B1Pyramidal optical flow tracker improvement
Publication Date: 2021.06.16 CONNAUGHT ELECTRONICS
  • EP3635684B1 patent drawingFigure 1~2
  • EP3635684B1 patent drawingFigure 3~5
  • EP3635684B1 patent drawingFigure 6~7

AI summary

The present invention refers to a method for performing optical flow estimation, comprising the steps of performing a detection of features (44) within a previous frame (42), providing an image pyramid (46) with representations (56) of different resolutions images of the previous frame (42) and a current frame (40), performing feature tracking between the previous frame (42) and the current frame (40) through the image pyramid (46), and calculating the optical flow between the previous frame (42) and the current frame (40) for the tracked features (44) based on the image pyramid (46), wherein the method comprises a step of performing a dynamic adaption of a number of pyramid layers (48, 50, 52, 54) of the image pyramid (46). The present invention also refers to a driving assistance system (12) comprising a camera (14, 16, 18) and a processing device (20), whereby the camera (14, 16, 18) provides subsequent frames and the processing device (20) is adapted to perform the above method using the frames provided from the camera (14, 16, 18). The present invention further refers to a vehicle (10) comprising an above driving assistance system (12).