Android – use area description file (ADF) in 3D reconstruction

I've been exploring ways to use Google's Tango project for 3D scanning and reconstruction

So far, some applications I have tried (such as Project Tango constructor and voxxlr) have done well in a short time range (I am happy to get suggestions for other potential scanning applications). The problem is that no matter what application I use, if I run long enough, the scanning results will accumulate a lot of drift, and eventually all contents will be aligned and damaged

Whenever I aim the device at a blank wall and other uncharacteristic space, or point the camera up to scan the ceiling, there will also be a great chance of drift. The device will temporarily lose its direction, thus destroying the alignment of future scanning. Anyway, it is a problem for me to let the device know where it is and what it points to

I know that some 3D scanning applications use area learning to some extent, because these applications ask me to allow area learning when starting the application. I think this is to help locate the device and stabilize its posture (if it is not accurate, please correct me)

From the applications I have tried, I have never obtained the option to load my own ADF. My understanding is that loading into the carefully learned feature rich ADF helps to better fix the posture of the device. Is there a reason for the lack of such applications that allow users to load their own ADF? Is it difficult / impossible? Has the current application made the best use of regional learning for localization, and can't any self recording ADF I provide do better?

I would appreciate any instructions / instructions on this topic - the method and effect of using ADF in 3D scanning and reconstruction are not clearly recorded. Finally, I am looking for a way to use tango for high-quality 3D scanning. If ADF is not required in the picture, that's good. If the answer is that I'm trying to complete an impossible task, I also want to know

If there is no ready-made solution, I am willing to try to deal with some cloud myself, although I think it is easier said than done

resolvent:

Unfortunately, Tango does not have any application to perform this operation at present, and you will need to develop your own application for this. In case you want to know how to perform this operation in the code, please follow the following steps: first, the learning mode of the application should be enabled. When we turn on the learning mode, the system will start recording ADF, This allows the application to view the existing areas it has been to. For each point cloud we save, we should also save the timestamp associated with the point

After walking around and collecting the points, we will need to call tangoservice from the API_ Saveareadescription function. This step optimizes each key pose saved in the system. After saving, we need to query again with the timestamp saved with the point cloud to optimize the pose. For this purpose, we use the function tangoservice_ Getposeattime. After completing this step, you will see that the point cloud has been set to the correct transformation, and the points will overlap

Just like reviewing the steps:

>Open the learning mode in tango configuration. > walk around and save the point cloud and the timestamp related to the point cloud. > call and save tangoservice_ SaveAreaDescription function. > save after completion, call TangoServcie_ Getposeattime to query the optimized pose according to the timestamp saved by the point cloud