Xiaobo Yang's research while affiliated with University of Electronic Science and Technology of China and other places

The extraction of features from LiDAR point clouds is a pivotal aspect of simultaneous localization and mapping (SLAM). However, in environments with weak features, such as parking structures and long corridors, the efficacy of current feature extraction methods is notably reduced. To overcome this limitation, a SLAM algorithm aimed at enhancing feature extraction performance is proposed. This algorithm commences by extracting edge and plane features from the original point clouds. Then, the plane points are categorized into ground points, ground-similarity points, and non-ground points, based on structural characteristics of the environment. Building upon this, a plane-point-based edge feature prediction method is proposed to generate virtual edge points of the environment, and the predicted points are then combined with the edge points to form the enhanced final edge features. In the map construction phase, a two-step Levenberg-Marquardt (LM) optimization process, guided by a novel feature regrouping strategy, is utilized to achieve high precision mapping. The performance of the algorithm is empirically validated in long corridor settings using the Livox Horizon LiDAR system. Comparative evaluations with Livox Loam and Fast LIO2 illustrate that our algorithm significantly improves feature extraction and localization accuracy.

This paper deals with the joint design of intra-inter agile pulses and Doppler filter banks to improve Doppler ambiguous target detectability in signal-dependent mainlobe clutter. Assuming that the target’s range and Doppler frequency (RDF) belong to an uncertainty set and that the clutter knowledge is partially known, a robust processing procedure based on a pulse compression matched filter and Doppler filter banks is proposed for the echo of the pulses with non-uniform pulse repetition interval (PRI). Then, the worst-case signal-to-interference-plus-noise ratio (SINR) at the output of all receive filters as the figure of merit is considered to be optimized along with both peak-to-average power ratio (PAR) constraint and energy constraint on the transmitted pulses. Utilizing an equivalent reformulation of the resultant non-convex maximin optimization problem, a maximum block improvement (MBI) framework that partitions the optimization variable into multiple blocks and updates only the block yielding the maximum increase of the objective function at each iteration, is developed to increase the worst-case SINR monotonically. Each block mainly invokes majorization-maximization (MM), sequential convex approximation (SCA) and entropy mirror descent (EMD) to obtain an enhanced solution. Two design strategies are also developed to seek valuable PRIs. Finally, at the analysis stage, the performance of the proposed robust technique is assessed, showing its capability for Doppler ambiguous target detection in a cluttered environment.

Building layout imaging (BLR) is a prominent research topic in the field of through-the-wall radar (TWR) and wireless perception currently. Inspired by computed tomography (CT), the transmit-receive separated dual-bistatic radar system utilizes electromagnetic (EM) wave transmission signals to perform BLR. However, existing researches significantly rely on signal frequency bandwidth resources. Moreover, the state-of-the-art researches only estimate the time delay (TD) information, thereby posing challenges in precisely discriminating between the direct path (DP) and multipaths. This paper refine the sparse signal reconstruction based angle of arrival (AOA) and TD super-resolution estimate algorithms under the condition of restricted broadband array signals. In accordance with this, the present study proposes a direct path (DP) identifying criterion with assistance of AOA, and obtain high-accuracy direct path time delay (DPTD) estimation. Further, with the high-accuracy DPTD estimation, this paper proposes a common material permittivities based iterative multimaterial BLR algorithm. The final numerical simulations and EM simulations verify the effectiveness of the proposed super-resolution algorithm and the improvement on multimaterial BLR.

In the design of resource-aware multi-target tracking (MTT) framework for multi-radar systems (MRSs), the estimated target parameters from the previous frame, such as distance, angle, and reflectivity, are generally utilized as prior information to guide resource scheduling at the current frame. However, achieving perfect estimation of these parameters is impossible in practice. Ignoring such uncertainty can result in unreliable solutions for resource scheduling. To address the uncertainty stemming from target parameter estimation during MTT, this paper proposes a robust strategy for joint radar-to-target assignment and power allocation (RJRAPA) in MRSs. At its core, the proposed method analyzes the probabilistic uncertainties of target parameters estimation within the confidence region and establishes the parameter uncertainty model. The posterior Cramér-Rao lower bound (PCRLB), considering parameter estimation uncertainty, is derived and adopted as the metric to evaluate target tracking accuracy, since it can provide a lower bound for the accuracy of the target state estimation. Then, based on the derived PCRLB, a robust resource scheduling problem is formulated with the goal of ensuring the optimal MTT accuracy across the whole confidence region of target parameters. The RJRAPA problem is shown to be non-convex with respect to the confidence regions of the target parameters. Thus, we present a convex relaxation approach for transforming the RJRAPA problem into a convex optimization problem with specific deterministic parameters. Numerical experiments that involve scenarios with parameter estimation uncertainties demonstrate the efficiency and robustness of our presented RJRAPA strategy.

As 3D point-cloud has the ability to present the contour of an object clearly, it provides more spatial information for person identification (PI) task. Aiming at the improvements on quality of point-cloud and distribution of features, an innovative treatment method for point-cloud and a novel network structure are investigated in this paper. Firstly, spatiotemporal feature of point-cloud is enhanced by implementing dual-stage density-based spatial clustering of applications with noise (DST-DBSCAN) method, which can filter most invalid points and decrease the sparsity of point-cloud. After that, the optimized point-cloud is input into neural network, which contains three parts for feature extraction, classification and feature optimization. Specifically, PointNet++ is adopted to extract features and realize PI recognition. In addition, an adversarial network is designed for optimizing feature distribution of point-clouds by encouraging the feature extractor of PointNet++ to generate features of the same person as similar as possible. Experimental results demonstrate that the proposed method can improve the accuracy by 3.77% than original PointNet++ network with raw data.

Localization of non-line-of-sight (NLOS) targets in the complex urban environment have attracted significant attention in recent years. However, the requirement for precise prior information about the environment is idealistic. It is challenging to know the environmental information in the blind area of vision in advance of practical applications. This paper proposes a joint estimation algorithm for building layout and target position in the L-shaped scene without any prior information. Specifically, a round-trip multipath propagation model is first developed for the cases of diffraction and multiple reflections. Then, the received echo signal is preprocessed with moving target identification (MTI), back-projection (BP) imaging, and image segmentation. In addition, the target points, which are screened by geometric association, are further matched and estimated by the multipath ghost’s hypothesis method, thus realizing the joint perceptual estimation of the building layout and the target position. Finally, electromagnetic (EM) simulations and experimental measurements are used to validate the effectiveness of the proposed algorithm.

An effective resource allocation strategy can maximize the remote sensing performance of radar systems, such as target detection and tracking. In this paper, two typical power allocation (PA) strategies are developed for the multi-target tracking (MTT) task in colocated MIMO (C-MIMO) radar systems with the consideration of the clutter. The multi-beam concept and the posterior PDF fusion are adopted by the C-MIMO radar system to obtain the global posterior distribution. Specifically, each radar generates multiple simultaneous beams with controllable power during each interval. To ensure that the limited system resources can be utilized effectively, the online PA scheme is implemented according to the prior knowledge predicted from the tracking cyclic recursive feedback results. The posterior Cramér-Rao lower bound (PCRLB) is derived by enumerating all possible target detection and false alarm occurrence cases, and is utilized as the tracking performance metric since it provides a more accurate lower bound on the target state estimation in clutter. Besides, to solve the computationally expensive problem of this PCRLB caused by enumeration operation, we propose a two-step approximate approach. Then, combined with the system resource configuration, two different types of resource optimization problems are designed, namely, performance maximization for a fixed power budget and direct resource minimization. These formulated PA problems are shown to be non-convex and non-linear. Therefore, we further propose a modified particle swarm optimization (MPSO) algorithm to solve these problems efficiently. Simulation results verify the superiority and effectiveness of the proposed PA strategies in terms of tracking performance in clutter.

In this paper, we consider a high-accuracy and low-complexity method for recognizing human activities behind wall. As the amount of information conveyed by data representation directly affects the recognition accuracy of the network, we construct the three-dimensional (3D) complex-valued feature for human activity recognition (HAR). In light of high network complexity introduced by 3D complex-valued data, we devise a low time and space complexity network named Convolutional Gated Recurrent Unit based on region-vectorization (RV-ConvGRU). Keystone Transform is utilized to process the radar echo and generate 3D complex-valued Range-Time-Doppler (RTD) data first, which provides high-frequency resolution and abundant feature information. Then, the real and imaginary parts of the complex-valued RTD are separately fed into a feature extraction module to comprehensively extract their respective features. Specifically, the real or imaginary part of the RTD is divided into multiple regions, which are then converted into regional vectors and reordered as channels to reduce the time and space complexity of the subsequent network. The reconfigured features are then input into the Convolutional Gated Recurrent Unit (ConvGRU) to extract global and temporal features, with the channel attention mechanism for feature selecting. The features of the real and imaginary parts are fused and then classified by the classifier finally. The experiments verify that the proposed method is effective, achieving the highest recognition accuracy of 99.23% with an input sequence of 1.44 seconds.