顾及动态物体感知的增强型视觉SLAM系统

传统的同步定位与制图（Simultaneous localization and mapping, SLAM）系统在复杂环境下工作时，无法分辨环境中的物体是否存在运动状态，图像中运动的物体可能导致特征关联错误，引起定位的不准确和地图构建的偏差。为了提高SLAM系统在动态环境下的鲁棒性和可靠性，本文提出了一种顾及动态物体感知的增强型视觉SLAM系统。首先，使用深度学习网络对每一帧图像的动态物体进行初始检测，然后使用多视图几何方法更加精细地判断目标检测无法确定的动态物体区域。通过剔除属于动态物体上的特征跟踪点，提高系统的鲁棒性。本文方法在公共数据集TUM和KITTI上进行了测试，结果表明在动态场景中定位结果的准确度有了明显提升，尤其在高动态序列中相对于原始算法的精度提升在92%以上。与其他顾及动态场景的SLAM系统相比，本文方法在保持精度优势的同时，提高了运行结果的稳定性和时间效率。     

POBI interpolation algorithm for CYGNSS near real time flood detection research: A case study of extreme precipitation events in Henan,China in 2021

In late 2016, NASA launched the first constellation of the global navigation satellite system reflectometry (GNSS-R) small satellites called the Cyclone Global Navigation Satellite System (CYGNSS). The stable data quality and continuous free availability of CYGNSS scientific data provided a new method for flood monitoring. However, owing to the pseudorandom distribution of CYGNSS data, researchers must always choose between high temporal resolution and high spatial resolution during the performance of flood monitoring based on CYGNSS data. For floods caused by extreme precipitation with sudden and short durations, the current flood mapping based on CYGNSS data cannot be updated in near real time. However, the near real time update of the flood distribution range is meaningful for postdisaster emergency response and rapid rescue. This study aimed to address this problem using a newly proposed spatial interpolation method based on previously observed behaviour (POBI). First, a method for calculating the surface reflectivity of the CYGNSS was introduced, followed by the principle of the POBI spatial interpolation method. The applicability of the POBI method in Henan Province, China, was then analysed, and by using the flood in Henan Province, China, in July 2021 as an example, the feasibility of CYGNSS near real time flood mapping based on the POBI method was evaluated. Based on the results, near real time and 3 km flood distribution monitoring results can be obtained using the proposed new method. The results were evaluated using MODIS (Moderate Resolution Imaging Spectroradiometer) images and compared with the observations of SMAP (Soil Moisture Active Passive) and GPM (Global Precipitation Measurement) in the same period. The results show that the flooded areas obtained by CYGNSS correspond to the inundated areas in MODIS images and are also in high agreement with the SMAP. In addition, CYGNSS allows for finer mapping and quantification of inundation areas and flood duration. Moreover, we also discussed the potential of CYGNSS to detect floods in shorter periods of time (a few hours) and did a preliminary evaluation using precipitation data from meteorological stations. The results are also highly consistent.