研究目的
To propose a novel and practical solution for the real-time indoor localization of autonomous driving in parking lots by extracting and enriching high-level landmarks (parking slots) with labels and introducing visual fiducial markers to improve precision.
研究成果
The proposed system effectively maps and localizes in indoor parking lots by detecting semantic landmarks (parking slots with IDs) and using visual fiducial markers for areas lacking slots. The robust optimization framework ensures correct data association and eliminates outliers. Future work aims to replace fiducial tags with other semantic clues for broader adaptability.
研究不足
The system relies on visual fiducial markers as an aid for loop closure, which may not be practical in all circumstances. The performance is affected by the quality of images, especially in dim or overexposed lighting conditions.
1：Experimental Design and Method Selection:
The system includes four fisheye cameras and one monocular camera for surround-view and front-view scenes, respectively. Parking slots are detected from top-view images fused from surround-view inputs. A CNN-based method is used for parking slot recognition. Visual fiducial markers (AprilTags) are introduced for areas lacking parking slots.
2：Sample Selection and Data Sources:
Datasets are collected from a parking lot in Jiading Campus, Tongji University, using the TiEV autonomous vehicle.
3：List of Experimental Equipment and Materials:
Four fisheye cameras, one monocular camera, IMU for vehicle speed and heading direction, and visual fiducial markers.
4：Experimental Procedures and Operational Workflow:
The vehicle is first operated by a human driver to initialize the parking map. Once the map stabilizes, the vehicle drives automatically following a pre-recorded trace based on real-time localization.
5：Data Analysis Methods:
A Graph-based optimization back-end with a Max-Mixture model is used for robust data association and outliers elimination.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
