Siamese KPConv: 3D multiple change detection from raw point clouds using deep learning

Gélis, Iris de; Lefèvre, Sébastien; Corpetti, Thomas

doi:10.1016/j.isprsjprs.2023.02.001

Cited by 23 publications

(21 citation statements)

References 63 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Therefore, we propose to study whether adding handcrafted features, and in particular a change-related feature, in Siamese KPConv [14] deep network influences the change segmentation results. As for hand-crafted features, we used the following ones related to:…”

Section: A Considering Hand-crafted Featuresmentioning

confidence: 99%

“…Among the different versions, the second one, with a LiDAR low density (around 0.5 points/m 2 ), is assessed here as it contains more classes of changes, and it relies on PCs more realistic than the first version of the dataset [14].…”

Section: A Datasetmentioning

confidence: 99%

“…Because of their specific characteristics, PCs are often rasterized into 2.5D Digital Surface Model (DSM) to be easily handled using traditional image processing methods. However, the rasterization implies a significant loss of information that can be prejudicial, thus more and more studies encourage designing methods able to deal with raw 3D PCs [13], [14].…”

mentioning

confidence: 99%

“…In a recent paper [14], we showed the possibilities brought by deep learning network to perform change detection and characterization into raw PCs. In particular, built upon 2D image change detection deep learning methods [15] and 3D PCs convolutions [16], the Siamese Kernel Point Convolution (KPConv) network outperforms traditional distance-based, DSM-based or machine learning-based methods on both real and simulated datasets for multiple change segmentation in urban areas.…”

mentioning

confidence: 99%

“…On the contrary, [19] propose to process the raw 3D PCs thanks to graph convolutions [20]. However, their method is designed for change classification task, i.e., retrieving changes at scene level as proposed by the Change3D [19] or Urban 3D Change Detection Classification (Urb3DCD-Cls) [14] datasets. This task is less precise than the change segmentation one, since it allows identifying only the main changed object in a scene without precisely localizing it.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Change Detection in Urban Point Clouds: An Experimental Comparison with Simulated 3D Datasets

2021

View full text Add to dashboard Cite

In the context of rapid urbanization, monitoring the evolution of cities is crucial. To do so, 3D change detection and characterization is of capital importance since, unlike 2D images, 3D data contain vertical information of utmost importance to monitoring city evolution (that occurs along both horizontal and vertical axes). Urban 3D change detection has thus received growing attention, and various methods have been published on the topic. Nevertheless, no quantitative comparison on a public dataset has been reported yet. This study presents an experimental comparison of six methods: three traditional (difference of DSMs, C2C and M3C2), one machine learning with hand-crafted features (a random forest model with a stability feature) and two deep learning (feed-forward and Siamese architectures). In order to compare these methods, we prepared five sub-datasets containing simulated pairs of 3D annotated point clouds with different characteristics: from high to low resolution, with various levels of noise. The methods have been tested on each sub-dataset for binary and multi-class segmentation. For supervised methods, we also assessed the transfer learning capacity and the influence of the training set size. The methods we used provide various kinds of results (2D pixels, 2D patches or 3D points), and each of them is impacted by the resolution of the PCs. However, while the performances of deep learning methods highly depend on the size of the training set, they seem to be less impacted by training on datasets with different characteristics. Oppositely, conventional machine learning methods exhibit stable results, even with smaller training sets, but embed low transfer learning capacities. While the main changes in our datasets were usually identified, there were still numerous instances of false detection, especially in dense urban areas, thereby calling for further development in this field. To assist such developments, we provide a public dataset composed of pairs of point clouds with different qualities together with their change-related annotations. This dataset was built with an original simulation tool which allows one to generate bi-temporal urban point clouds under various conditions.

show abstract

Section: A Considering Hand-crafted Featuresmentioning

confidence: 99%

Section: A Datasetmentioning

confidence: 99%

mentioning

confidence: 99%

mentioning

confidence: 99%

mentioning

confidence: 99%

See 3 more Smart Citations