Detection of Geometric Temporal Changes in Point Clouds

Abstract

Detecting geometric changes between two 3D captures of the same location performed at different moments is a critical operation for all systems requiring a precise segmentation between change and no‐change regions. Such application scenarios include 3D surface reconstruction, environment monitoring, natural events management and forensic science. Unfortunately, typical 3D scanning setups cannot provide any one‐to‐one mapping between measured samples in static regions: in particular, both extrinsic and intrinsic sensor parameters may vary over time while sensor noise and outliers additionally corrupt the data. In this paper, we adopt a multi‐scale approach to robustly tackle these issues. Starting from two point clouds, we first remove outliers using a probabilistic operator. Then, we detect the actual change using the implicit surface defined by the point clouds under a Growing Least Square reconstruction that, compared to the classical proximity measure, offers a more robust change/no‐change characterization near the temporal intersection of the scans and in the areas exhibiting different sampling density and direction. The resulting classification is enhanced with a spatial reasoning step to solve critical geometric configurations that are common in man‐made environments. We validate our approach on a synthetic test case and on a collection of real data sets acquired using commodity hardware. Finally, we show how 3D reconstruction benefits from the resulting precise change/no‐change segmentation.

Enhanced Visualization of Detected 3D Geometric Differences

“…Obviously, the choice of the change detection algorithm can partially influence the effectiveness of the method, even if the principles behind it still remain valid. In the released visualization tool, we add a demo with two different versions of a Gargoyle where the change field was computed with different algorithms: the algorithm from [PCBS16] and the Hausdorff distance [CRS98] normalized with respect to the maximum distance. The relative change maps are shown in Figure .…”

“…To overcome this problem, we interpolate the alpha parameters and in a neighbourhood of the change threshold d . For this purpose, we require a change field with a continuous and smooth radial distribution around the real changes, like the field computed with a simple Hausdorff distance or with [PCBS16]. In particular, for all the pixels with at least one change value and in the range , we interpolate the alpha blending parameters using the following function (Figure b): The obtained result is a smoother transition near the boundary of the binary segmentation that makes less abrupt and more pleasant the final colour interpolation (Figure ).…”

“…Viewpoints of the real datasets used in the first and second user studies (Sections and ). Each viewpoint shows the two times, the relative change maps computed with [PCBS16] (blue = no‐change, red = change), the parameter d for the change/no‐change classification and the size of the two meshes in millions of triangles.…”

“…To detect the changes between the input meshes, we use a variant of the method presented in [PCBS16]. The algorithm computes a multi‐scale growing least square (GLS) [MGB*12] descriptor, independently for each mesh, with an adaptive sub‐sampling of the volume occupied by the models using an octree.…”

“…An enhanced and interactive visualization of temporal data that improve the perception and understanding of the changes in the scene is useful for several monitoring applications, like the environmental disaster prevention and management, the monitoring of construction sites, archaeological excavations or urban environment, the preservation and presentation of Cultural Heritage artefacts in the time. In the last years, several solutions have been proposed to solve the geometric change detection problem for different scenarios (3D reconstruction [YSL*14], urban growth analysis [TBP13] and natural events management [LFM*13]) using different input data (image datasets [PM07], 3D models and photos [TBP13] and only 3D models [PCBS16]). On the other side, the effective visualization of temporal data has been focused on time‐varying volumetric data [WWS03, KBH*10, CRY11] and on videos [BDG15].…”