Advancements in Synthetic Video Generation for Autonomous Driving
Based on approaches from previous works , their advantages and limitations , we propose a novel approach which suggests fusion of Generative Adversarial Neural Network , Deep Learning , and Image processing . The proposed system aims to build an environment that will provide a onestop solution for different AI ( Artificial Intelligence ) rendering , scenario generation and future video prediction . We also propose various possibilities 9 through which an end-user can generate realistic data from our system . For the testing of generated outputs , we have taken evaluation scores based on Fréchet Inception Distance ( FID ) 10 and Kanade-Lucas-Tomasi ( KLT ) 11 scores to reduce human intervention .
The organization of this report is as follows : Chapter 2 includes a survey , Chapter 3 includes proposals and contributions , Chapter 4 includes methodology , and Chapter 5 includes results and conclusions .
2 MOTIVATION
2.1 SURVEY
Research was conducted previously for image-to-image translation 12 , 13 , 14 . Video-to-video synthesis aspects have also been explored in the recent past [ 4 ]–[ 6 ], but have not been observed to be highly effective . Our work aims to explore the different possibilities from an end-user perspective of the application . It also aims to improve the performance of the latest work done in this area to develop a system that can help in the faster development of advanced driver assistance systems ( ADAS ).
K . K . Patel [ 2 ] explores the possibilities of reducing the reality gap in autonomous vehicles development , for which simulator-based environments are used to generate multiple scenarios of different environments . CARLA 15 , an open-source simulation environment for autonomous driving research , aims to support the development , training , and validation of various autonomous driving modes .
9 https :// arxiv . org / abs / 1609.01326
10 https :// arxiv . org / abs / 1706.08500
11 https :// ieeexplore . ieee . org / document / 323794
12 https :// arxiv . org / abs / 1711.11585
13 https :// arxiv . org / abs / 1903.07291
14 https :// arxiv . org / abs / 1611.07004
15 https :// arxiv . org / abs / 1711.03938 Journal of Innovation 109