Optical Flow Matrix Generation from DVS Timestamp Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current optical flow calculation methods based on Dynamic Vision Sensor (DVS) data are complex and time-consuming, failing to meet real-time performance requirements due to the need for local hyperplane fitting and extensive neighborhood searches.
Innovation Solution
The method generates a timestamp matrix from event data flow and scans it in predetermined directions to determine intermediate points, calculating distances and gradient directions, thereby eliminating the need for local plane-fitting calculations and improving real-time performance by directly generating an optical flow matrix through element traversal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If local hyperplane fitting and neighborhood search are performed for each pixel point, then measurement precision of optical flow is improved, but device complexity and computation time increase significantly
Solution Approach 1:
The patent extracts only the essential information needed for optical flow calculation from the event data, organizing it into timestamp matrices. Instead of performing complex local hyperplane fitting for each pixel, the method extracts gradient information and performs simplified calculations on the organized matrix data, thereby reducing computational complexity while maintaining measurement precision.
Solution Approach 2:
The calculation process is segmented into distinct stages: event data collection, timestamp matrix generation, gradient calculation, and optical flow computation. This segmentation allows each stage to be optimized independently, reducing overall complexity while preserving the precision benefits of neighborhood analysis.
2Measurement precision
If local hyperplane fitting is performed for each pixel point, then optical flow calculation precision is improved, but loss of time increases due to extensive computation
Solution Approach 1:
The patent performs preliminary organization of event data into timestamp matrices before the actual optical flow calculation. This preliminary action pre-processes the data into a format that enables faster gradient computation and optical flow derivation, reducing the time required for the main calculation while maintaining precision through preserved neighborhood relationships.
Solution Approach 2:
The patent replaces the mechanical process of local hyperplane fitting with a more efficient computational approach using timestamp matrices and gradient calculations. This substitution maintains the essential measurement precision while dramatically reducing computation time by avoiding iterative optimization procedures.
3Adaptability or versatility
If event data is processed using traditional frame-based methods, then compatibility with existing systems is maintained, but productivity decreases due to computational burden
Solution Approach 1:
The patent changes the fundamental parameter representation from frame-based images to event-based timestamp matrices. This parameter change enables the system to process optical flow at event rates rather than frame rates, dramatically improving productivity while maintaining adaptability through the flexible matrix structure that can accommodate different sensor configurations.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure provides an optical flow calculation method for a computing device, including: acquiring an event data flow with a predetermined duration from a DVS, the event data flow including a coordinate position and a timestamp of a triggered event; generating a timestamp matrix in accordance with the coordinate position and the timestamp of the triggered event; scanning elements in the timestamp matrix in a predetermined scanning direction, so as to determine at least one intermediate point in each element in accordance with a value and a gradient of the element in the predetermined scanning direction; and calculating a distance between adjacent intermediate points and a gradient direction, and generating an optical flow matrix in accordance with a calculation result. The present disclosure further provides the computing device.