Spike-Based Vision for Autonomous Driving Scenarios

Abstract

Autonomous driving is an important research direction in computer vision which has broad application prospects. Pure vision perception schemes have significant research value in autonomous driving scenarios. Different from traditional cameras, spike vision sensor offers imaging speeds over a thousand times faster than traditional cameras, possess advantages such as high temporal resolution, high dynamic range, low data redundancy, and low power consumption. This study focuses on autonomous driving scenarios, introducing the imaging principles, perception capabilities, and advantages of the spike camera. Besides, focusing on visual tasks related to autonomous driving, this study elaborates on the principles and methods of spike-based image/video reconstruction, discusses the approach to image enhancement based on sensor fusion with spike cameras,and provides a detailed description of the algorithms and technical routes for motion optical flow estimation, object recognition, detection, segmentation, and tracking, and deep estimation of three-dimensional scenes based on spike cameras. It also summarizes the development of the spike camera data and systems. At last, it analyzes the challenges, potential solutions, and future directions for spike vision research. Spike cameras and their algorithms and systems hold great potentials in the field of autonomous driving and represent one of the future research directions in computer vision.

Keywords

• spike-based vision
• spike camera
• autonomous driving
• artificial intelligence