RESUMO
Autonomous systems need to localize and track surrounding objects in 3D space for safe motion planning. As a result, 3D multi-object tracking (MOT) plays a vital role in autonomous navigation. Most MOT methods use a tracking-by-detection pipeline, which includes both the object detection and data association tasks. However, many approaches detect objects in 2D RGB sequences for tracking, which lacks reliability when localizing objects in 3D space. Furthermore, it is still challenging to learn discriminative features for temporally consistent detection in different frames, and the affinity matrix is typically learned from independent object features without considering the feature interaction between detected objects in the different frames. To settle these problems, we first employ a joint feature extractor to fuse the appearance feature and the motion feature captured from 2D RGB images and 3D point clouds, and then we propose a novel convolutional operation, named RelationConv, to better exploit the correlation between each pair of objects in the adjacent frames and learn a deep affinity matrix for further data association. We finally provide extensive evaluation to reveal that our proposed model achieves state-of-the-art performance on the KITTI tracking benchmark.
RESUMO
In this paper, we present challenges and achievements in development and use of a compact ultrasonic Phased Array (PA) module with signal processing and imaging technology for autonomous non-destructive evaluation of composite aerospace structures. We analyse two different sets of ultrasonic scan data, acquired from 5 MHz and 10 MHz PA transducers. Although higher frequency transducers promise higher axial (depth) resolution in PA imaging, we face several signal processing challenges to detect defects in composite specimens at 10 MHz. One of the challenges is the presence of multiple echoes at the boundary of the composite layers called structural noise. Here, we propose a wavelet transform-based algorithm that is able to detect and characterize defects (depth, size, and shape in 3D plots). This algorithm uses a smart thresholding technique based on the extracted statistical mean and standard deviation of the structural noise. Finally, we use the proposed algorithm to detect and characterize defects in a standard calibration specimen and validate the results by comparing to the designed depth information.
RESUMO
Nonlinear modal analysis is a demanding yet imperative task to rigorously address real-life situations where the dynamics involved clearly exceed the limits of linear approximation. The specific case of geometric nonlinearities, where the effects induced by the second and higher-order terms in the strain-displacement relationship cannot be neglected, is of great significance for structural engineering in most of its fields of application-aerospace, civil construction, mechanical systems, and so on. However, this nonlinear behaviour is strongly affected by even small changes in stiffness or mass, e.g., by applying physically-attached sensors to the structure of interest. Indeed, the sensors placement introduces a certain amount of geometric hardening and mass variation, which becomes relevant for very flexible structures. The effects of mass loading, while highly recognised to be much larger in the nonlinear domain than in its linear counterpart, have seldom been explored experimentally. In this context, the aim of this paper is to perform a noncontact, full-field nonlinear investigation of the very light and very flexible XB-1 air wing prototype aluminum spar, applying the well-known resonance decay method. Video processing in general, and a high-speed, optical target tracking technique in particular, are proposed for this purpose; the methodology can be easily extended to any slender beam-like or plate-like element. Obtained results have been used to describe the first nonlinear normal mode of the spar in both unloaded and sensors-loaded conditions by means of their respective backbone curves. Noticeable changes were encountered between the two conditions when the structure undergoes large-amplitude flexural vibrations.
RESUMO
In large civil aircraft manufacturing, a time-consuming post-production process is the non-destructive inspection of wing panels. This work aims to address this challenge and improve the defects' detection by performing automated aerial inspection using a small off-the-shelf multirotor. The UAV is equipped with a wide field-of-view camera and an ultraviolet torch for implementing non-invasive imaging inspection. In particular, the UAV is programmed to perform the complete mission and stream video, in real-time, to the ground control station where the defects' detection algorithm is executed. The proposed platform was mathematically modelled in MATLAB/SIMULINK in order to assess the behaviour of the system using a path following method during the aircraft wing inspection. In addition, two defect detection algorithms were implemented and tested on a dataset containing images obtained during inspection at Airbus facilities. The results show that for the current dataset the proposed methods can identify all the images containing defects.