DZDonglin Zhu

Donglin Zhu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Donglin Zhu has a total of 17 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Gaussian landmarks tracking-based real-time splatting reconstruction model

    by Donglin Zhu, Zhongli Wang, Xiaoyang Fan, Miao Chen, Jiuyu Chen


  • OrthSynergy: Orthogonal feature-based synergy descriptor for loop closure detection in large-scale scenes

    by Donglin Zhu, Zhongli Wang, Miao Chen, Tengfei Lu, Le Gao, Xiaotao Shao, Yan Shen


  • An Efficient Geometric Points-Guided Generalized Exponential Splatting

    by Donglin Zhu, Zhongli Wang, Xin Jiang

    This paper presents a novel Generalized Exponential Splatting framework for scene reconstruction. Unlike neural radiance fields (NeRF), 3D Gaussian Splatting (3DGS) offers rapid rendering and high fidelity. However, existing 3DGS frameworks integrating RGB-D SLAM face challenges in system efficiency and aliasing artifacts. Inspired by geometric features, we first categorize the space into three regions—edge, planar, and non-planar—based on their information richness. The edge extraction is realized through an energy function, while planar and non-planar regions are segmented via a fast segmentation algorithm proposed in this paper. Secondly, a sparseness function is established to screen key points within the these regions to build a geometric points-guided Generalized Exponential Splatting model (GP-GES), strategically placing primitives to reduce redundancy. Finally, by integrating a 3D low-pass filter, a smooth GP-GES model is further built, ensuring that the frequency of the primitives adheres to the Nyquist-Shannon theorem, to mitigate high-frequency artifacts. Extensive experiments show that our method reduces the map size by 3.7× to 6× without sacrificing rendering quality, thereby accelerating the computational efficiency of the system. In addition, the segmentation algorithm has a fast detection speed, with a frame rate exceeding 90Hz.

  • MHTraj: A Multi-Domain Hybrid Graph Neural Network With Causal-Spatial Modeling for Multi-Agent Trajectory Prediction

    by Jiuyu Chen, Chunxiao Jia, Wei Xie, Donglin Zhu, Xiaotao Shao, Zhongli Wang

    As a crucial part of social behavior modeling, multi-agent trajectory prediction with dynamic interactions has become one of the core technologies for autonomous systems. One of the main research streams is to leverage relational graph Transformers to learn interactions among agents, as this method demonstrates superior capability in capturing implicit features and modeling complex relationships. However, such researches fall short in efficiently utilizing multi-domain features, consequently being constrained by spatial attention layers that hinder accuracy improvements. To address the above limitations and achieve a more lightweight map-free architecture, this work introduces a Multi-domain Hybrid Graph Transformer (MHTraj) for predicting multi-modal future trajectories. The core insight of the method lies in its well-designed encoder, which enhances inputs via a frequency-domain mixer and employs causal convolution attention to capture global temporal information. Furthermore, a graph Transformer is incorporated to aggregate spatial relationships, thereby yielding the final hybrid features. Extensive experiments on 4 real-world datasets (NBA, ETH/UCY, SDD, nuScenes) show that the proposed method achieves competitive state-of-the-art (SOTA) performances, and meanwhile reducing the parameter count by nearly 50% on the ETH/UCY datasets.


  • PMF-SLAM: Pose-Guided and Multiscale Feature Interaction-Based Semantic SLAM for Autonomous Wheel Loader

    by Donglin Zhu, Zhongli Wang, Tengfei Lu, Xin Jiang

    In a dynamic environment, semantic information can assist the SLAM system in eliminating dynamic point interference. However, most three-dimensional semantic segmentation methods are computationally expensive, which also have low segmentation accuracy for both distant and small objects. We propose a PMF-SLAM method to fully exploit the interaction between 3D semantic segmentation and SLAM and achieve efficient scene perception. The PMF-SLAM system includes three parts: MSF-SegNet model, Interactive SLAM module, Pose-Guiding segmentation module. To improve the accuracy of distant and small objects, MSF-SegNet merges point-wise global features and voxel-wise local features from two branches by a designed symmetrical sparse convolution structure. In the Interactive SLAM module, the coarse-to-fine registration method based on semantic information completes the estimation of pose. To implement the interaction between Segmentation and SLAM, the Pose-Guiding segmentation module was built to assist the segmentation thread in improving inference efficiency and ensuring segmentation consistency over time. Extensive experiments including both local experiment and nuScenes dataset test have been conducted to validate the performance of the proposed method. Our method achieves better accuracy than multiple segmentation algorithms, significantly improving the segmentation performance of distant and small objects. And the trajectory estimation accuracy is better than multiple SLAM algorithms. Code is available at https://github.com/haroldgt/MSF-SegNet.


  • PairCon-SLAM: Distributed, Online, and Real-Time RGBD-SLAM in Large Scenarios

    by Donglin Zhu, Guanghui Xu, Xiaoting Wang, Xiaogang Liu, Dewei Tian

    This article proposed a PairCon-simultaneous localization and mapping algorithm to construct dense maps of large-scale scenarios in real time, which is operated in an integral platform containing two personal computers (PCs). This platform can operate the mapping thread independently in a PC to guarantee sufficient memory resources. In this context, the synchronous visualization of 3-D maps can be achieved in such a platform. In contrast, traditional algorithms are limited by the short operation distance, under the requirement of simultaneous data collection and 3-D maps construction in a PC. In addition, the memory resource provided by a single PC is limited, which restricts the constructing scale of maps. Thus, we use a separate PC to construct maps independently to relieve the distance constraint and exploit the socket method to conduct the transmission of data with the point cloud. Meanwhile, we introduce the ReBlur algorithm into the semi-direct method to reduce the error accumulation of the odometer in the tracking thread, which improves the robustness performance. In addition, the method combining the memory management and DBow2 algorithm is adopted to improve the accuracy of loop detection. In the considered system, the quality of maps and the performance of the odometer are evaluated by ICL-NUIM and the datasets, such as TUM, DIODE, and so on, respectively. Finally, under the simulation environment of AirSim and Gazebo, we construct maps based on the image data of other scenarios, which is used to show the quality of the construction.


  • Adaptive Multi-Sensor Data Fusion Positioning Algorithm Based On Long Short Term Memory Neural Networks

    by Jie Zhou, Guanghui Xu, Donglin Zhu, Enbiao Di

    In the indoor Non-Line-Of-Sight (NLOS) positioning scenes, improve the accuracy and robustness of positioning can be the main challenge. To solve this challenge, an adaptive algorithm based on long short-term memory neural networks (LSTM) modified multi-sensor data fusion positioning is proposed by us. First, we compare the single sensor and multi-sensor data fusion methods to show the superiority of multi-sensor data fusion, which determined our next research. Then, in multi-sensor data fusion scene, in order to reduce inertial measurement unit (IMU) cumulative errors and ultra-wide band (UWB) measured errors, LSTM is used for data modeling and error prediction based on the measured values of IMU and UWB. Finally, on the foundation of the prediction results, the accuracy of IMU and UWB fusion positioning is enhanced by adjusting the coefficients. Experimental results reveal that the algorithm of adaptive adjusting the coefficients by LSTM reduces the root mean square error (RMSE) and average error contrasted with the Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF).