Abstract
As the acquisition of laser range measurements such as those from light detection and ranging (LiDAR) sensors requires a considerable amount of time, to design an effective sampling algorithm is a critical task in numerous laser range applications. The state-of-the-art adaptive methods such as two-step sampling are highly effective at handling less complex scenes such as indoor environments with a moderately low sampling rate. However, their performance is relatively low in complex on-road environments, particularly when the sampling rate of the measuring equipment is low. To address this problem, this paper proposes a region-of-interest (ROI)-based sampling algorithm in on-road environments for autonomous driving. With the aid of fast and accurate road and object detection algorithms, particularly those based on convolutional neural networks, the proposed sampling algorithm utilizes the semantic information and effectively distributes samples in the road, object, and background areas. The experimental results demonstrate that the proposed algorithm significantly reduces the mean-absolute-error in the object area by at most 52.8% compared to two-step sampling; moreover, it achieves robust reconstruction quality even at a very low sampling rate of 1%.
Highlights
In recent years, autonomous driving has become an emerging trend
To mimic the complex natural sensing system of humans, a vehicle is installed with different types of sensors such as grayscale/color cameras, inertial and GPS navigation sensors, radio detection and ranging (RADAR), and light detection and ranging (LiDAR) sensors [2], [3]
This study addresses the problem of finding sampling locations for LiDAR scanners to minimize the reconstruction error in an entire scene or a specific region-of-interest (ROI) for a given sampling budget
Summary
Autonomous driving has become an emerging trend. From the sensing aspect, 3-D cameras such as RGB-D cameras and light detection and ranging (LiDAR) sensors are becoming more affordable, enabling both academic studies and industrial (commercial) applications, such as self-driving cars employing video analytics on LiDAR captured data for path planning as well as obstacle detection [1]. Tasks in autonomous driving is the generation of a local map of objects (i.e., road, vehicles, and pedestrians) surrounding a car. This task directly relies on the depth sensing technologies. A. SAMPLING MODEL Let x ∈ RN be an N ×1 vector representing a depth map of an entire scene in an FOV of a capturing device such as a LiDAR. A LiDAR sensor cannot acquire data for all the locations in the FOV so that it reconstructs the depth map of the entire FOV from the sampled data. Let M denote the number of samples from which a capturing device is capable of acquiring data. For mathematical formulation, let {1, . . . , N } represent the set of the indexes that correspond to the locations of the entire FOV and {i1, . . . , iM} represent the set of the indexes that correspond to the sampling locations among {1, . . . , N }
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.