http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Pseudo-LiDAR를 활용한 자율주행 영상인식 시스템의 3D 객체 검출 성능향상 연구
이철우(Cheolwoo Lee),백초혜(Chohye Baek),나희연(Heeyeon Nah) 한국자동차공학회 2022 한국자동차공학회 학술대회 및 전시회 Vol.2022 No.11
With the rise of artificial intelligence (AI) and the success of various deep neural network (DNN) applications, the autonomous driving has gained significant interest as one of promising research fields in both industry and academia. Autonomous driving technologies are largely divided into three stages which are perception, decision and control, and AI is making great research achievements in the perception. So far, the LiDAR sensor has been considered as a primary sensing media since it provides accurate depth information. Nevertheless, researchers have sought the alternative of the LiDAR due to the high cost and power consumption, as well as its unattractive design. In this work, we present an alternative, cost-effective and highly accurate 3D object detection mechanism built upon a simple stereo camera sensor, while providing performance comparable to the one based on LiDAR sensor. By converting images into point cloud called Pseudo-LiDAR and using it, then we can achieve 3.6x greater accuracy than image based algorithms. In terms of speed, we can achieve 18x faster inference time than MV3D or F-PointNet. In conclusion, an accurate and cost-effective object detector can be made by combining Complex-YOLO and the Pseudo-LiDAR method.