Multiple Camera and 3D Point Cloud Problems

[mmuna]

Hi,

Our project requires multiple cameras (SR300) to capture a fixed target. We encountered some problems during the development as follow:

1. Set the order of multiple cameras.

In the enumeration of cameras, we can get many parameters from the DeviceInfo(SDK 2016 R3) object. The parameter "didx" is described in the document is "The zero-based device index if there are multiple devices on the system". But we can not determine which camera's "didx" will be 1, which will be 2. Is there any way to determine and modify the camera "didx", or orther id numbers?

2. Splice the point cloud from multiple cameras (at different viewing angles) into a complete 3D model.

We learned that the PCL contains some algorithms for point cloud registration, but the development language we used was C# . Is there any algorithm or method in RealSense SDK can be used to do this( point cloud registration & Camera calibration)? Is there any other good way to deal with it?

3. Triangulation of the 3D point cloud.

The SDK 3D Scan module reconstructs the shape and appearance of stationary objects from a sequence of images, which converted into a 3D triangle mesh (e.g. PLY). Did the SDK provide methods to convert the 3D point cloud from the raw stream(not use 3D scan algorithm module) into a 3D model file? In addition, is there any other way to do triangulation of the 3D point cloud, especially using C# language?

We hope to receive your valuable suggestions.

Thanks!

[idata]

Hi Arthur,

Thanks for your interest in the Intel® RealSense Platform. We appreciate all the information provided and we would like to investigate a little bit more your inquires and as soon as we find useful information we'll let you know. We'll appreciate your patience during the meantime. Regards,-Yermi

[idata]

Hi Arthur,

Thank you for your patience and interest in Intel RealSense technology. Below are the answers to your questions.1. The order of camera enumeration is not customizable and the camera IDs are not modifiable.2. The RealSense SDK does not provide a native method for point cloud registration. For further information you may want to see these articles: https://software.intel.com/en-us/articles/intel-realsense-technology-and-the-point-cloud https://software.intel.com/en-us/articles/intel-realsense-technology-and-the-point-cloud and https://github.com/IDLER1229/Registration https://github.com/IDLER1229/Registration.3. The SDK does not natively provide a method for doing this but it can be done programmatically. See this article: https://software.intel.com/en-us/articles/perpetual-computing-generating-3d-from-depth-data https://software.intel.com/en-us/articles/perpetual-computing-generating-3d-from-depth-data.I hope this information is helpful.Best regards,Jesus GarciaIntel Customer Support

[idata]

I am also working to implement a 3d scanner using multiple cameras. I wonder if this can be done using the RealSense SDK. I looked at the example 3d scanning code, and it looks like mayb it is possible to modify it to aquire images from multiple cameras and then reconstruct all the images together.. But would it work? Or is there aother way of doing this?

Eyal

[idata]

Hello Eyal,

The RealSense SDK for Windows does not allow for multi-camera capture but the RealSense Cross-Platform API (also known as librealsense) does have this capability. You can download it and get more information from: https://github.com/IntelRealSense/librealsense. Librealsense does not provide scanning MW so you will have to develop that layer yourself.There are other issues you will have to solve that are much more difficult. First, having multiple SR300 cameras pointed at the same object causes very bad destructive IR interference in depth. Many people have tried to get this to work by turning the lasers on and off then trying to synchronize the frames. Very awkward. This use case is not supported.The R200 camera may be better suited but at close ranges we have also noticed some destructive interference when two cameras are pointed to the object simultaneously. You will have to test this in your environment.Regards,Jesus GarciaIntel Customer Support