Real-time 3D Perception of Scene with Monocular Camera

Shadi Saleh,Shanmugapriyan Manoharan,Wolfram Hardt

doi:10.14464/ess.v7i2.436

Abstract

   Depth is a vital prerequisite for the fulfillment of various tasks such as perception, navigation, and planning. Estimating depth using only a single image is a challenging task since the analytic mapping is not available between the intensity image and its depth where the features cue of the context is usually absent in the single image. Furthermore, most current researchers rely on the supervised Learning approach to handle depth estimation. Therefore, the demand for recorded ground truth depth is important at the training time, which is actually tricky and costly. This study presents two approaches (unsupervised learning and semi-supervised learning) to learn the depth information using only a single RGB-image. The main objective of depth estimation is to extract a representation of the spatial structure of the environment and to restore the 3D shape and visual appearance of objects in imagery.  

Highlights

Understanding of visual scenes is a vital component of many applications of Artificial Intelligence, ranging from autonomous vehicles to the navigation of household robots and even automatic annotation of imagery for the blind
DEPTH ESTIMATION TECHNIQUES Estimating depth using only a single RGB frame is often considered an ill-posed and inherently ambiguous challenge, because the analytic mapping is not available between the intensity image and its depth where the features cue of the context is usually absent in the single image
Self-supervised Depth Estimation: This method is based on the Structure from Motion (SFM) framework

Summary

INTRODUCTION

Understanding of visual scenes is a vital component of many applications of Artificial Intelligence, ranging from autonomous vehicles to the navigation of household robots and even automatic annotation of imagery for the blind. Accurate depth perception for moving and stationary obstacles can provide more knowledge about the environment, which helps autonomous vehicles to take valid action in critical situations based on 3D perception information. This problem in stereo vision sensors is resolved by computing a disparity image from the stereo image pair to extract depth information as described [8]. The alternative solution for obtaining the depth would be to employ range sensors such as Lidar or radar These are naturally highly accurate sensors that provide highly precise depth measurements. Reduce model complexity and the number of samples required for training in order to maximize the accuracy and speed of deep learning algorithms

RESEARCH QUESTIONS AND OBJECTIVES

DEPTH ESTIMATION TECHNIQUES

Depth Estimation based on Semi-Supervised Learning

CONCLUSION & FUTURE SCOPE