EFFECTS OF 2D VIDEO PROJECTION ON OBJECT IDENTIFICATION IN SPHERICAL VIDEO FOR AN AUTONOMOUS VEHICLE APPLICATION

Loading...
Thumbnail Image

Date

2022-05

Journal Title

Journal ISSN

Volume Title

Publisher

The Ohio State University

Research Projects

Organizational Units

Journal Issue

Abstract

Due to the growing population density, the demand for efficient personal navigation has been increasing rapidly. Part of this field is covered by interest in autonomous vehicles (AV's). To meet the demand for this technology, many research efforts have been directed toward studying computer vision and navigation algorithms. Additionally, the use of computer vision has numerous other growing uses of importance – some of which include imaging and mapping. Computer vision is also being used within the technologies of growing popularity, altered reality (AR) and virtual reality (VR). Past research studies and current uses of computer vision often use many different sensors, including a variety of 2D camera setups. Within the field however, there is significantly less usage of spherical video. In this regard, this study aims to characterize the advantages and disadvantages of the use of a single spherical camera for object identification in an autonomous driving scenario as well as the resulting 2D video projections from the source, which often impose distortion on the video content. First, camera placement on the vehicle's roof - above the driver - was picked to encompass a viewing field prioritizing roadway obstacles, pedestrians, and street signs. Second, this setup was used to capture spherical video during varying conditions while on trips around The Ohio State University. The video was then formatted into four different 2D videos using a variety of projection techniques and run under a popular convolutional neural network object detector, called YOLO. Results from this study will advance our understanding of computer vision and how it is used effectively. While primarily for autonomous vehicle scenario, the resulting data provides an insight on the usability of spherical video for image recognition in general.

Description

Keywords

Spherical Video, Object Identification, Sphere to 2D Video Projection, Computer Vision, Autonomous Driving

Citation