Search

Abstract Accurate positioning and navigation play a vital role in vehicle-related applications, such as autonomous driving and precision agriculture. With the rapid development of Global Navigation Satellite Systems (GNSS), Precise Point Positioning (PPP) technique, as a global positioning solution, has been widely applied due to its convenient operation. Nevertheless, the performance of PPP is severely affected by signal interference, especially in GNSS-challenged environments. Inertial Navigation System (INS) aided GNSS can significantly improve the continuity and accuracy of navigation in harsh environments, but suffers from degradation during GNSS outages. LiDAR (Laser Imaging, Detection, and Ranging)-Inertial Odometry (LIO), which has performed well in local navigation, can restrain the divergence of Inertial Measurement Units (IMU). However, in long-range navigation, error accumulation is inevitable if no external aids are applied. To improve vehicle navigation performance, we proposed a tightly coupled GNSS PPP/INS/LiDAR (GIL) integration method, which tightly integrates the raw measurements from multi-GNSS PPP, Micro-Electro-Mechanical System (MEMS)-IMU, and LiDAR to achieve high-accuracy and reliable navigation in urban environments. Several experiments were conducted to evaluate this method. The results indicate that in comparison with the multi-GNSS PPP/INS tightly coupled solution the positioning Root-Mean-Square Errors (RMSEs) of the proposed GIL method have the improvements of 63.0%, 51.3%, and 62.2% in east, north, and vertical components, respectively. The GIL method can achieve decimeter-level positioning accuracy in GNSS partly-blocked environment (i.e., the environment with GNSS signals partly-blocked) and meter-level positioning accuracy in GNSS difficult environment (i.e., the environment with GNSS hardly used). Besides, the accuracy of velocity and attitude estimation can also be enhanced with the GIL method.

The interpretation of deep learning network is an important part in understanding the convolutional neural networks (CNNs). As an exploratory research, this article explored the interpretation method in 3-D point cloud deep learning networks, for the purpose of evaluating the performance of convolution functions in 3-D point cloud CNNs. Specifically, a 3-D point cloud classification network with two branches is used as the interpretation network in two aspects; 1) information entropy is introduced to diagnose the internal representation in the middle layer of CNN; and 2) the external consistency of convolution function is measured by per-point classification accuracy with class activation mapping technique. Four typical convolution functions are tested by the interpretation network on ModelNet40 dataset and the experimental results demonstrate that the proposed evaluation method is reliable. Feature transformation ability and feature recognition ability of convolution functions are extracted by visualization evaluation and proposed measurable metrics evaluation.