Dynamic Vision Sensor Characterization Using Frame-to-Event Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods struggle to characterize dynamic vision sensors (DVS) due to non-linear relationships between manufacturing processes, temperature, and scene illumination, making it difficult to extract accurate parameters.
Innovation Solution
A method involving converting frames from an image sensor into events using conversion parameters, determining differences between converted and detected events, and adjusting these parameters to minimize discrepancies, utilizing a DVS model that emulates the DVS's physical and operating characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional analysis methods are used to characterize DVS, then the analysis process is simple, but the measurement precision of DVS parameters is insufficient due to non-linear relationships between manufacturing processes, temperature, and scene illumination
Solution Approach 1:
The patent creates a computational model that copies the DVS response characteristics to generate synthetic events from frame sequences. This model replicates the non-linear behavior of actual DVS pixels under varying temperature and illumination conditions, enabling accurate parameter extraction without requiring complex physical measurement setups.
Solution Approach 2:
The patent introduces frame sequences from a conventional image sensor as an intermediary to bridge the gap between easily measurable frame data and the difficult-to-measure DVS event output. By converting frames to synthetic events through the computational model, the system indirectly characterizes DVS parameters that would otherwise require complex direct measurement.
2Measurement precision
If the number of parameters characterizing the DVS is increased to capture non-linear relationships, then the measurement precision improves, but the difficulty of detecting and measuring these parameters increases
Solution Approach 1:
The patent employs an optimization algorithm that uses feedback from the comparison between synthetic events (generated from frames using current parameter estimates) and actual DVS events to iteratively refine parameter values. The feedback loop continuously adjusts parameters to minimize the difference between modeled and measured event streams, enabling accurate extraction of multiple non-linear parameters.
Solution Approach 2:
The patent performs preliminary conversion of frame sequences into synthetic events using initial parameter estimates before the optimization process begins. This preliminary action creates a baseline comparison dataset that guides the subsequent parameter refinement, making the measurement of multiple non-linear parameters more tractable.
3Productivity
If the DVS operating parameters are adjusted to optimize performance, then the productivity of machine vision systems improves, but the device complexity increases due to multiple underlying physical and operating parameters
Solution Approach 1:
The patent systematically varies operating parameters such as threshold voltage, integration time, and bias conditions during the characterization process to map their effects on DVS response. By establishing the relationships between these parameters and actual performance metrics through the computational model, the system enables optimized parameter selection without requiring manual trial-and-error adjustment of multiple complex variables.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables precise characterization of DVS parameters, facilitating quality verification during manufacturing and optimizing operating settings for improved performance in machine vision systems.
Implementation Method 1
a DVS detects so-called 'events' when an illumination and, thus, a light stimulus of a pixel of the DVS changes
Data Source
AI summary
The present disclosure relates to a method for characterizing a dynamic vision sensor, DVS, comprising converting frames of a scene generated by an image sensor into events of the scene based on conversion parameters characterizing the DVS. The method further comprises determining a difference between the converted events and events corresponding to the scene detected by the DVS. The method further comprises adjusting the conversion parameters to reduce the difference between the converted and the detected events.