Event-Based Block-Matching Optical Flow Algorithm

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional optical flow algorithms are not suitable for event-based vision sensors like DVS, as they are frame-based and struggle with dense textured inputs, and existing methods fail to provide efficient real-time computation on embedded platforms.

Innovation Solution

A new event-based block-matching algorithm that accumulates change events in three time slice frames and uses a distance metric to find the most similar search block, implemented both in software and on FPGA, allowing for efficient computation of optical flow with improved accuracy and speed.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional frame-based optical flow algorithms are used with event-based vision sensors, then the algorithms can be directly applied to standard vision data, but they fail to work effectively with sparse event-based inputs and dense textured scenes

Engineering Contradiction:
Improvecompatibility with event-based sensorsVSAvoidaccuracy on dense textured input
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent transforms event-based sparse inputs into frame-based representations by accumulating events over time intervals, converting the data format to match conventional optical flow algorithm requirements while preserving the event-based sensor's advantages

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent introduces an intermediate representation layer that converts event streams into pseudo-frame structures, allowing conventional algorithms to process event-based data without direct modification while maintaining accuracy on textured inputs

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If block matching with dense image blocks is used to improve optical flow accuracy, then measurement precision is improved, but computational cost increases making real-time embedded implementation difficult

Engineering Contradiction:
Improveoptical flow accuracyVSAvoidreal-time computation speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent divides the image into blocks and processes them independently through parallel hardware pipelines, enabling accurate block matching while maintaining real-time performance through concurrent computation

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent replaces sequential software-based block matching with parallel hardware circuit implementation on FPGA, achieving both high accuracy and real-time performance by substituting software computation with dedicated hardware logic

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

3Productivity

If existing event-based optical flow methods are implemented on embedded platforms, then real-time processing is achieved, but they are not designed for logic circuit implementation and lack efficiency

Engineering Contradiction:
Improvereal-time processing capabilityVSAvoidimplementation complexity on logic circuits
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent implements optical flow computation directly in hardware logic circuits on FPGA, replacing software algorithms with dedicated hardware pipelines that achieve real-time performance with optimized resource utilization

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

Solution Approach 2:

The patent designs a dynamic hardware architecture that adapts processing resources based on input event rates and computational requirements, optimizing performance for different scene complexities while maintaining real-time operation

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3631756B1Block-matching optical flow and stereo vision for dynamic vision sensors
Publication Date: 2022.01.19 UNIVERSITY OF ZURICH
  • EP3631756B1 patent drawingFigure 1
  • EP3631756B1 patent drawingFigure 2
  • EP3631756B1 patent drawingFigure 3~4

AI summary

The invention relates to computation of optical flow using event-based vision sensors.