Adaptive Slicing Method of the Spatiotemporal Event Stream Obtained from a Dynamic Vision Sensor.

Area of Science:

• Computer Vision
• Sensor Technology
• Signal Processing

Background:

• Dynamic Vision Sensors (DVS) excel in high dynamic range, temporal resolution, and low power consumption.
• DVS technology offers advantages over traditional cameras, particularly in challenging computer vision scenarios.
• The asynchronous spatiotemporal event stream from DVS presents visualization and compatibility challenges for existing algorithms.

Purpose of the Study:

• To develop a novel adaptive slicing method for processing dynamic vision sensor spatiotemporal event streams.
• To enhance the compatibility of DVS data with traditional image processing techniques.
• To ensure complete object information and eliminate motion blur in the processed event data.

Main Methods:

• An adaptive slicing method was developed to segment the spatiotemporal event stream.
• The method generates slices containing comprehensive object information without motion blur.
• Slices can be processed using event-based algorithms or by conversion into virtual frames for traditional algorithms.

Main Results:

• The proposed slicing method effectively processes DVS data, yielding motion-blur-free object information.
• Validation using public and custom datasets confirmed the method's efficacy.
• Object information entropy in the slices closely matched ideal entropy, with a difference of less than 1%.

Conclusions:

• The adaptive slicing method successfully addresses the visualization and compatibility issues of DVS event streams.
• This technique enables the use of DVS data with both event-based and traditional computer vision algorithms.
• The method holds significant potential for advancing computer vision in demanding applications.