Corner Detection in Dynamic Vision Sensors Using Event Density
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing corner detection algorithms for Dynamic Vision Sensors (DVS) are time-consuming and do not fully utilize the event-driven hardware advantages, as they convert event frames into bitmap images for processing, which fails to accurately reflect the chronological order of events and information about moving targets.
Innovation Solution
A corner detection method that directly identifies suitable points in the DVS event frame for optical flow calculation by dividing the event frame into areas, determining the number of events in each area, and selecting corners based on these counts and a preset maximum number, thereby avoiding bitmap image conversion and considering the time sequence of events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If DVS event frame is converted into bitmap image for corner detection, then traditional bitmap-based algorithms can be applied, but processing time increases significantly and hardware advantages of DVS are not utilized
Solution Approach 1:
The patent extracts only the essential corner information directly from the DVS event frame without converting to bitmap image. It identifies corner candidates by analyzing event density and temporal patterns in specific regions, extracting only the necessary corner points while discarding unnecessary image conversion steps, thus reducing processing time while maintaining accuracy
Solution Approach 2:
The patent replaces the mechanical image conversion process with a direct event-based computational approach. Instead of converting events to bitmap pixels and then applying traditional algorithms, it substitutes this mechanical transformation with a mathematical model that directly processes event streams using temporal and spatial statistics to detect corners, eliminating the intermediate conversion step
2Measurement precision
If bitmap image conversion is performed for corner detection, then existing algorithms can be used, but the chronological order of DVS events is not preserved
Solution Approach 1:
The patent changes the parameter representation from spatial bitmap pixels to temporal event sequences. It transforms the corner detection problem from analyzing static image intensities to analyzing dynamic event arrival times and frequencies, using temporal parameters such as event density over time and inter-event intervals to identify corners while preserving chronological information
Solution Approach 2:
The patent adds the temporal dimension to corner detection by analyzing events across time rather than only in spatial space. It constructs a four-dimensional event space (x, y, t, polarity) and detects corners by examining patterns in this extended dimensionality, particularly using the time dimension to distinguish true corners from transient features, thus preserving temporal information that would be lost in bitmap conversion
3Productivity
If static bitmap image is used for corner detection, then simple algorithms can be applied, but moving target information is not accurately reflected
Solution Approach 1:
The patent transforms the static corner detection approach into a dynamic one by continuously analyzing the evolving pattern of events over time. It tracks how event density and distribution change temporally at each spatial location, identifying corners based on dynamic patterns of event accumulation and dissipation rather than static image features, thus accurately reflecting moving target information while maintaining processing efficiency
Solution Approach 2:
The patent performs preliminary temporal analysis of event patterns before final corner determination. It pre-processes the event stream by computing temporal statistics such as event rates and accumulation patterns in advance, preparing dynamic features that indicate potential corners before the actual detection decision is made, thereby enabling accurate moving target detection without sacrificing processing speed
Data Source
AI summary
A corner detection method includes: acquiring a dynamic vision sensor (DVS) event frame, the DVS event frame including at least one event and being pre-divided into a plurality of areas; determining, of the at least one event, a number of events in each of the plurality of areas; detecting corners in each of the plurality of areas based on the number of events in each of the plurality of areas and a preset maximum number of corners; and obtaining a plurality of corners of the DVS event frame based on the detected corners.