APL-VIO: Real-Time Visual-Inertial Odometry Using Adaptive Point-Line Features
Wangyan Li, Lifeng Wu, Xuewen Xiao, Wei Xiao, Jie Bao
University of Shanghai for Science and Technology Shanghai Dianji University UNSW Sydney
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
Recently, the use of line features to improve the performance ofvisual-inertial odometry(VIO) has received significant attention. However, most of them are limited to monocular scenarios or utilize the line structure information from a single camera. Besides, the low success rate and considerable computational effort of the line feature tracking make it challenging to meet the efficiency and accuracy requirements of the VIO system. To address these challenges, this work proposes a real-time VIO system, which utilizes adaptive point-line features, namely APL-VIO, to fully exploit the line structure information from all fields of view, whether using stereo or monocular cameras. Specifically, we present anadaptive line optical flow tracker(ALOFT) that efficiently tracks line features while guaranteeing their quality and quantity of line features. Furthermore, we introduce a novel line constraint reconstruction method that provides line constraints across multiple frames, enhancing the accuracy of the VIO system and reducing the redundancy in the reconstructed line map. Next, thePlücker coordinatesandorthonormal representationtechniques are introduced to parameterize space lines for transformation and optimization, respectively. In what follows, the line reprojection residuals are modeled, along with point andinertial measurement unit(IMU) measurement residuals, and are utilized to obtain the optimal VIO system state estimates through a sliding window optimization (SWO). Finally, the proposed APL-VIO is compared with other VIO methods in datasets and experiments, respectively. Both comparisons show that the proposed APL-VIO exhibits satisfactory real-time performance and achieves the highest average accuracy.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
计算机 / AIRobotics and Sensor-Based Localization
Inertial Sensor and Navigation · Advanced Vision and Imaging