Search

Simultaneously estimating the kinematic state and extent of extended targets is a nonlinear and high-dimensional problem. While the extended Kalman filter (EKF) is widely employed to achieve this goal, it may not be sufficient for mobility targets. To address this issue, this paper first proposes to embed unscented Kalman filter (UKF) into Gaussian process regression (GPR) since the superiority of UKF to high nonlinear. Furthermore, given the widely-existed environment with time-varying noise, it is crucial to study the change of measurement noise covariance caused by time-varying noise for high-precision tracking of extended targets. However, traditional UKF filter considers measurement noise covariance as constant value. To this end, an adaptive unscented Kalman filter (AUKF) algorithm combining with GPR model (GPR-AUKF) is proposed to address the issue. Specifically, the GPR-AUKF algorithm is built based on expectation maximization (EM) algorithm to track the target state and covariance, and which updates the measurement noise covariance in real-time. Experimental results show that GPR-AUKF is more accurate and robust than other methods for tracking extended targets.

The segmentation of airborne lidar scanning (ALS) point clouds is one of the basic tasks in remote sensing field. The existed learning-based methods have acquired satisfactory performance, however, the training process requires a large amount of labeled data, which limits their wide application to some extent. Weakly-supervised learning of point clouds could achieve competitive results but far from perfect compared to fully supervised methods. In this article, we propose a weakly-supervised method for semantic segmentation of ALS point clouds based on learning of line and plane points. In addition, as basic feature units widely existing in large-scale scenes, both line and plane features could provide certain supervisory signals for training. Specifically, we extract line and plane points through two hand-crafted methods, which achieve offline line and plane point extraction. The extracted line and plane points are considered as additional supervisory signals, and consequently two auxiliary tasks could be constructed based on prediction of line and plane points, respectively. Furthermore, a multistage learning framework is proposed to embed the two auxiliary tasks into existing network, which makes the feature learning of network in the line-plane-object way. Moreover, we introduce an improved pseudolabel generation strategy based on prediction-consistency of contiguous models, which could generate high-quality pseudolabels so that to improve the performance of network. Extensive experiments conducted on three ALS point cloud datasets, i.e., ISPRS, LASDU, and DFC2019, demonstrate that the proposed method achieves considerable gains upon the baseline with respect to F1 score and overall accuracy under 0.1$\%$, 0.5$\%$, and 0.05$\%$ weak labels.

In the case of a weak signal from a photon counting lidar and strong noise from the solar background, the signal is completely submerged by noise, potentially resulting in the appearance of multiple peaks in the denoising algorithm of photon counting entropy. Consequently, a clear distinction between the signal and noise may become challenging, leading to significant fluctuation in the ranging error. To solve this problem, this paper proposes an improved offset parameter optimization algorithm under the framework of photon counting entropy, aiming to effectively eliminate peak interference caused by noise and enhancing ranging accuracy. The algorithm includes two aspects. First, we introduce the solar irradiance prediction of an MLP network and least squares linear conversion to accurately estimate the noise rate of the solar background noise. Then, we propose the offset parameter optimization method to effectively mitigate the interference caused by noise. In simulation and experimental analyses, the ranging error of our proposed method is within 5 and 30 cm, respectively. Compared with the denoising method of photon counting entropy, the average ranging error is increased by 81.99% and 73.76%. Furthermore, compared to other anti-noise methods, it exhibits superior ranging capability.

Abstract The motion averaging problem arises when recovering the motions for a set of scans in three‐dimensional (3D) reconstruction. Motion averaging is taken as an optimisation problem, whose solution can be obtained via matrix decomposition in principal component analysis. The latest motion averaging methods utilize the Cauchy kernel function or overlap rate to construct the weighted matrix in the cost function, yielding weighted low‐rank sparse matrix decomposition solutions. Nevertheless, the over‐reliance on the initial value and high computational complexity remain two major problems. Here, we aim to improve the motion averaging performance via a fast and robust method based on the angle constraints of multiple scans. The additional term representing the angle constraint is added into the original cost function. By doing this, two benefits are facilitated: 1) the iteration times for the optimisation process are reduced; and 2) it effectively mitigates the over‐reliance on the initial value and thus caters for outliers and random noise of initial motions. Compared with state‐of‐the‐art motion averaging methods, it is indicated that the proposed method is more efficient and robust for numerical simulations of using both simulated motions and real 3D scanning data, given a large range of angle constraint values.