A Time-Related Voxel Representation Method for Event Camera

摘要

The event camera's high temporal resolution and dynamic range make it an optimal tool for detecting and tracking high-speed moving objects. However, the asynchronous and sparse output event data of the event camera limits the use of traditional image processing methods in its application. In this paper, a time-related voxel representation method is proposed as a solution to overcome the challenge of sparse and asynchronous output event data in the event camera. This method efficiently represents the discrete event data in a compact voxel form, arranged based on their temporal order and polarity. By using this representation method, we can apply existing image processing techniques to the event data while reducing information loss. The proposed event representation method achieved a remarkable classification accuracy of 0.998 and 0.967 for the N-MNIST datasets and NCARS datasets, respectively. This outperforms existing event representation methods, demonstrating the high effectiveness of the proposed method for classification tasks. These results highlight the potential of the time-related voxel representation method to enhance the performance of event cameras and their application in various fields, including robotics, autonomous vehicles, and surveillance.