This article provides a comprehensive survey of deep learning applications for autonomous vehicle object detection and scene perception. It begins with an introduction to self-driving cars, deep learning, and computer vision, before discussing existing powerful deep learning libraries and their role in deep learning growth. It then examines techniques that address the image perception issues in real-time driving, and evaluates recent implementations and tests conducted on self-driving cars. Finally, it provides several recommendations for further research into the use of deep learning for safe, human-intervention-free driving.
Key Points
Cameras are image sensors that operate on RGB values, capturing infrared visual data and offering high resolution information.
RADARs are ultrasonic and highly reliable, providing higher resolution and accuracy, and working well in extreme weather, but having low resolution.
LiDARs are expensive but provide extremely accurate depth information and 360 degrees visibility, making them useful for 3D ground truth data in driving environments.
Potential solutions to real time road data analysis: multimodal sensor fusion, road scene analysis in adversarial weather conditions, and polarimetric image analysis for object detection in autonomous driving. Self driving cars must implement data driven learning models as it is impossible to calculate all the "if-then-else" cases.
Citation
Abhishek Gupta, Alagan Anpalagan, Ling Guan, Ahmed Shaharyar Khwaja,
Deep learning for object detection and scene perception in self-driving cars: Survey, challenges, and open issues,
Array,
Volume 10,
2021,
100057,
ISSN 2590-0056,
https://doi.org/10.1016/j.array.2021.100057.
Title
Deep learning for object detection and scene perception in self-driving cars
URL
https://www.sciencedirect.com/science/article/pii/S2590005621000059
Summary
This article provides a comprehensive survey of deep learning applications for autonomous vehicle object detection and scene perception. It begins with an introduction to self-driving cars, deep learning, and computer vision, before discussing existing powerful deep learning libraries and their role in deep learning growth. It then examines techniques that address the image perception issues in real-time driving, and evaluates recent implementations and tests conducted on self-driving cars. Finally, it provides several recommendations for further research into the use of deep learning for safe, human-intervention-free driving.
Key Points
Citation
Abhishek Gupta, Alagan Anpalagan, Ling Guan, Ahmed Shaharyar Khwaja, Deep learning for object detection and scene perception in self-driving cars: Survey, challenges, and open issues, Array, Volume 10, 2021, 100057, ISSN 2590-0056, https://doi.org/10.1016/j.array.2021.100057.
Repo link