COMPETITION

Did you know that only 65% of the imagery of the Construction sites could be alligned using commercial software ( i.e. MetaShape and RealityCapture)? This is not so surprising if you know the many obstacles Structure-from-Motion pipelines have to overcome to allign the imagery below.

37.PNG

In this competition, we want to spark innovation for more robust alignment procedures. This includes the calculation of both the interior and exterior camera parameters. As an example, you can take a look at what information is stored from a software like RealityCapture

38.PNG

Given these parameters, we can accurately position the cameras in the construction coordinate system. Using the GEOMAPI API, we can easily import some imagery and point clouds. For instance, in the example below, we load a set of images from a MetaShape xml file.

import geomapi
from geomapi import tools as tl
from geomapi.utils import geometryutils as gmu

imgNodes=tl.xml_to_nodes(path_to_metashape_xml)

las= laspy.read(las_path)
pcdNode=PointCloudNode(name='myPointCloud', resource=gmu.las_to_pcd(las))

And we can visualize these resources using Open3D and some placeholder geometries for the imagery. Note that merging geometries before sending them to the visualizer significantly speeds up the rendering.

joinedImages=gmu.join_geometries([gmu.generate_visual_cone_from_image(n.cartesianTransform, height =1).paint_uniform_color([1,0,0]) for n in imgNodes])
o3d.visualization.draw_geometries([joinedImages]+[pcdNode.resource])

Now the COMPETITION is to align as many of the resources as possible. All you have to do is report your percentage of properly aligned imagery on the complete dataset.

2.PNG