This week’s installment covers a tool that gathered a bit of dust on my shelf but proved to be an invaluable resource when modeling scene evidence this week, sparking a renewed interest and appreciation.

​RealityCapture [RC] is a powerful photogrammetry tool that excels at creating 3D models from a large photoset. As discussed here, it can be used for processing drone imagery, but it can also be used for creating detailed point clouds and meshes of any object with notable texture.

Upload, say, 100 photographs and the program uses computer vision techniques to identify unique features in every photograph. From there, RC estimates and refines all camera locations and creates a sparse point cloud based on the matched features. With all camera positions solved, dense point clouds and meshes can be created by triangulating depth information for each pixel.

From that description, you may see why it’s not great when working with large, featureless, panels like fenders, hoods, quarter panels, etc. However, if you have a rock, a roadway, a shoe, or even a dirt bike, the technology performs remarkably well.

The rendering below shows a point cloud of a Honda XR400 generated with 60 photos. There are a few gaps on the larger, smooth sections, but overall, the algorithm did a good job. As a side note, Recon-3D works in a very similar manner, but also receives depth-information from the iPhone’s/iPad's LiDAR sensor, hence its ability to handle objects with minimal surface detail.

 
 

The RC learning curve is a little steep, but this section of a PowerPoint from my photogrammetry class walks through the process step-by-step. If you want to give it a shot, here’s a link to download the photos used to create the Honda above. RC is free if your firm's gross revenue is less than $1M and $1,250 per year otherwise.

I recommend giving it a whack, it’s a tool worth exploring!

Lou Peck
​Lightpoint | Axiom