•We provide a unique classification-based 3D change detection dataset from a complex street environment. There are no other open 3D point cloud datasets released for our purpose.•We evaluate ...different algorithms on the dataset and help finding solutions for 3D point cloud change detection tasks.•The results show that the proposed siamese graph convolutional networks (SiamGCN) are good at extracting representative geometric features and can hereby outperform compared algorithms on the 3D change detection dataset.
Display omitted
The rapid development of 3D acquisition devices enables us to collect billions of points in a few hours. However, the analysis of the output data is a challenging task, especially in the field of 3D point cloud change detection. In this Shape Retrieval Challenge (SHREC) track, we provide a street-scene dataset for 3D point cloud change detection. The dataset consists of 866 3D object pairs in year 2016 and 2020 from 78 large-scale street scene 3D point clouds. Our goal is to detect the changes from multi-temporal point clouds in a complex street environment.
We compare three methods on this benchmark, with one handcrafted (PoChaDeHH) and the other two learning-based (HGI-CD and SiamGCN). The results show that the handcrafted algorithm has balanced performance over all classes, while learning-based methods achieve overwhelming performance but suffer from the class-imbalanced problem and may fail on minority classes. The randomized oversampling metric applied in SiamGCN can alleviate this problem. Also, different siamese network architecture in HGI-CD and SiamGCN contribute to the designing of a network for the 3D change detection task.
We propose in this paper a new approach to solve the decision problem of robot-following. Different from the existing single policy model, we propose a multipolicy model, which can change the ...following policy in time according to the scene. The value of this paper is to obtain a multipolicy robot-following model by the self-learning method, which is used to improve the safety, efficiency, and stability of robot-following in the complex environments. Empirical investigation on a number of datasets reveals that overall, the proposed approach tends to have superior out-of-sample performance when compared to alternative robot-following decision methods. The performance of the model has been improved by about 2 times in situations where there are few obstacles and about 6 times in situations where there are lots of obstacles.
Localization and navigation are the two most important tasks for mobile robots, which require an up-to-date and accurate map. However, to detect map changes from crowdsourced data is a challenging ...task, especially from billions of points collected by 3D acquisition devices. Collecting 3D data often requires expensive data acquisition equipment and there are limited data sources to evaluate point cloud change detection. To address these issues, in this Shape Retrieval Challenge (SHREC) track, we provide a city-scene dataset with real and synthesized data to detect 3D point cloud change. The dataset consists of 866 pairs of object changes from 78 city-scene 3D point clouds collected by LiDAR and 845 pairs of object changes from 100 city-scene 3D point clouds generated by a high-fidelity simulator.
We compare three methods on this benchmark. Evaluation results show that data-driven methods are the current trend in 3D point cloud change detection. Besides, the siamese network architecture is helpful to detect changes in our dataset. We hope this benchmark and comparative evaluation results will further enrich and boost the research of point cloud change detection and its applications.
Display omitted
•A representative dataset providing high-quality annotated 3D point clouds, which contains rich urban objects and covers a wide range of change detection challenges.•The dataset fills the vacancy in the point cloud change detection task.•A comprehensive evaluation of point cloud change detection approaches. We organize the Shape Retrieval Challenge (SHREC) benchmark on point cloud change detection.•Different evaluation metrics are used to compare the proposed methods based on our datasets.
The effectiveness of crossties in six sets of large-scale reinforced concrete short columns under monotonic axial compression is evaluated experimentally. The primary variables were type of crosstie ...engagement, type of crosstie, and crosstie placement offset. It was observed that minor placement offset up to 20 mm between hooks of crossties and longitudinal steel bars does not bring unfavorable consequences in confinement. The crosstie hook engagement of both hoop and longitudinal steel bar is better than that engagement of the longitudinal steel bar alone, but it is difficult in installation. Crossties composed of two straight end-180-deg hook steel bars give the best performance among the types of crosstie and it is convenient to facilitate the fabrication of reinforcing cages. However, the effectiveness under cyclic loading needs further study.
Frequent trajectory pattern mining is an important spatiotemporal data mining problem with broad applications. However, it is also a difficult problem due to the approximate nature of spatial ...trajectory locations. Most of the previously developed frequent trajectory pattern mining methods explore a crisp space partition approach 8,10 to alleviate the spatial approximation concern. However, this approach may cause the sharp boundary problem that spatially close trajectory locations may fall into different partitioned regions, and eventually result in failure of finding meaningful trajectory patterns. In this paper, we propose a flexible vague space partition approach to solve the sharp boundary problem. In this approach, the spatial plane is divided into a set of vague grid cells, and trajectory locations are transformed into neighboring vague grid cells by a distance-based membership function. Based on two classical sequential mining algorithms, the PrefixSpan and GSP algorithms, we propose two efficient trajectory pattern mining algorithms, called VTPM-PrefixSpan and VTPM-GSP, to mine the transformed trajectory sequences with time interval constraints. A comprehensive performance study on both synthetic and real datasets shows that the VTPM-PrefixSpan algorithm outperforms the VTPM-GSP algorithm in both effectiveness and scalability.
In this paper, we present a novel robust and fast object tracker called spatial kernel phase correlation based Tracker (SPC). Compared with classical correlation tracking which occupies all spectrums ...(including both phase spectrum and magnitude spectrum) in frequency domain, our SPC tracker only adopts the phase spectrum by implementing using phase correlation filter to estimate the object׳s translation. Thanks to circulant structure and kernel trick, we can implement dense sampling in order to train a high-quality phase correlation filter. Meanwhile, SPC learns the object׳s spatial context model by using new spatial response distribution, achieving superior performance. Given all these elaborate configurations, SPC is more robust to noise and cluster, and achieves more competitive performance in visual tracking. The framework of SPC can be briefly summarized as: firstly, phase correlation filter is well trained with all subwindows and is convoluted with a new image patch; then, the object׳s translation is calculated by maximizing spatial response; finally, to adapt to changing object, phase correlation filter is updated by reliable image patches. Tracking performance is evaluated by Peak-to-Sidelobe Ratio (PSR), aiming to resolve drifting problem by adaptive model updating. Owing to Fast Fourier Transform (FFT), the proposed tracker can track the object at about 50frames/s. Numerical experiments demonstrate the proposed algorithm performs favorably against several state-of-the-art trackers in speed, accuracy and robustness.
The increasing availability of tracking devices bring larger amounts of trajectories representing people's moving location histories. In this paper, we aimed to mine closed frequent patterns in ...moving trajectory database. Such closed frequent patterns can help us to understand general mobile behaviors in compact representation. In this work, we first presented a conception of spatiotemporal region of interesting (STROI) to capture the attribute of moving trajectory in spatial and temporal dimensions. Second, based on the set of STROIs distributing in given geospatial region, we transformed trajectory data into STROI element sequence data at different time slice with respect to corresponding STROIs. Third, we modified the closed sequence pattern mining algorithm CloSpan to adapt to closed moving trajectory pattern discovery. Finally, the approaches are then validated by a range of synthetic data sets to evaluate the usefulness and efficiency.
•Provide a large-scale 3D street-scene point cloud dataset for 3D semantic segmentation.•Evaluate different algorithms on the dataset and help finding solutions for large-scale 3D point cloud ...processing. We have five algorithms under evaluation with one based on handcrafted detectors, one based on 3D-to-2D projection learning, and the other three being end-to-end learning-based methods.•The results show that point-set based end-to-end learning methods can learn representative features directly from 3D points and performs better than handcrafted methods.
Display omitted
Scene understanding of large-scale 3D point clouds of an outer space is still a challenging task. Compared with simulated 3D point clouds, the raw data from LiDAR scanners consist of tremendous points returned from all possible reflective objects and they are usually non-uniformly distributed. Therefore, its cost-effective to develop a solution for learning from raw large-scale 3D point clouds. In this track, we provide large-scale 3D point clouds of street scenes for the semantic segmentation task. The data set consists of 80 samples with 60 for training and 20 for testing. Each sample with over 2 million points represents a street scene and includes a couple of objects. There are five meaningful classes: building, car, ground, pole and vegetation. We aim at localizing and segmenting semantic objects from these large-scale 3D point clouds. Four groups contributed their results with different methods. The results show that learning-based methods are the trend and one of them achieves the best performance on both Overall Accuracy and mean Intersection over Union. Next to the learning-based methods, the combination of hand-crafted detectors are also reliable and rank second among comparison algorithms.
PDF
