Multi-sensor-based detection and tracking of moving objects for relative position estimation in autonomous driving condi
- PDF / 3,438,928 Bytes
- 23 Pages / 439.37 x 666.142 pts Page_size
- 11 Downloads / 165 Views
Multi‑sensor‑based detection and tracking of moving objects for relative position estimation in autonomous driving conditions Jinwoo Kim1 · Yonggeon Choi1 · MyungWook Park1 · Sangwoo Lee1 · Sunghoon Kim1
© Springer Science+Business Media, LLC, part of Springer Nature 2019
Abstract Moving object detection (MOD) technology was combined to include detection, tracking and classification which provides information such as the local and global position estimation and velocity from around objects in real time at least 15 fps. To operate an autonomous driving vehicle on real roads, a multi-sensor-based object detection and classification module should carry out simultaneously processing in the autonomous system for safe driving. Additionally, the object detection results must have high-speed processing performance in a limited HW platform for autonomous vehicles. To solve this problem, we used the Redmon in DARKNET-based (https://pjreddie.com/darknet/yolo) deep learning method to modify a detector that obtains the local position estimation in real time. The aim of this study was to get the local position information of a moving object by fusing the information from multi-cameras and one RADAR. Thus, we made a fusion server to synchronize and converse the information of multi-objects from multi-sensors on our autonomous vehicle. In this paper, we introduce a method to solve the local position estimation that recognizes the around view which includes the long-, middle- and short-range view. We also describe a method to solve the problem caused by a steep slope and a curving road condition while driving. Additionally, we introduce the results of our proposed MOD-based detection and tracking estimation to achieve a license for autonomous driving in KOREA. Keywords Moving object detection · Deep learning · Local position estimation · Sensor fusion
* Jinwoo Kim [email protected] Extended author information available on the last page of the article
13
Vol.:(0123456789)
J. Kim et al.
1 Introduction As deep learning-based technologies continue to develop, the trends in vehicle technology have also rapidly changed to advanced technology including the fusion of vision and sensor technologies. In particular, detection and classification technology has developed rapidly for autonomous self-driving systems. Many areas of autonomous driving recognition require detection, tracking, motion estimation and prediction in diverse conditions because self-driving systems are different from advanced driver assistance systems (ADASs) which just provide the driver assistance. Because self-driving systems will have to deal with diverse environments including dangerous events on real roads such as during an accident, they need to be safe. Actually, the big data should be converged to recognize the driving circumstance more and more. It is important to make an acquisition and definition for the driving data from anywhere and anything that helps in driving. Recently, the researches and ideas are introduced to solve big data that the na
Data Loading...
