Best Practice: How to choose a suitable Object Target

The Wikitude SDK requires reference maps from the objects that should be recognized. Wikitude uses standard videos as the source for extracting relevant information and creating map files. This document describes what to consider when recording the video for an optimal input for the map creation.  

In order to ensure that a high quality 3D map of the object can be created based on the video provided you need to take care of two things, choosing a suitable object and recording a proper video. In this article we emphasize on the first one but we also share information regarding the record process.

1. Finding an ideal object is highly important in order to be able to track it correctly later on.

When you choose an object you need to ensure that:

  • The object should not cast any clearly visible shadows
  • Reflective surfaces should be avoided
  • There should not be any disturbances of other objects in the scene
  • The object should not change its original form. For example, scanning a shoe could be a problem since the form of the shoe could change. If the strings for example are tied or not could result in a different object.
  • Typical objects that work well are small-scale objects ranging up to 1,5 meters

Examples

The first most common mistake when choosing an object is objects with the same texture. The following videos represent some bad examples of objects due to repetitive pattern with not clear edges and corners.


The second most common mistake is  objects whose surface is reflective.

Although the following object could be used as a good example, its reflective surface makes it a poor choice.

Finally, picking objects that are too small is not good for the tracking process. Although the following object is considered a good example, its size makes the tracking almost impossible. Even if you would zoom in during the video recording, when you would try to track the real size of the object the results would not be satisfactory.

If you are not sure if the object you picked is good you can also create the Object Target and have a look at the Cloud Points of your object. For instance, the following object resulted in this image, which is missing a lot of Cloud points.

Screen Shot 2017-07-17 at 13.28.19.png


2. Having selected the ideal object, the next step would be to record a video of this object. During this process what you need to check is that:

  • There needs to be sufficient lighting when recording the video - ideally a very homogenous indirect light.
  • Smooth movement of the camera is recommended. Fast motions can influence the quality of the map negatively.
  • It needs to be ensured that the object is recorded from all perspectives and distances that it should be recognized in the end.

The best process when creating a video is:

  1. Start recording the x-axis of the object.
  2. Continue with a full rotation of the video. The movement needs to be smooth and slow.
  3. Scan the object from above

170629_WT_ObjectRecording_Infografik_03.jpg

Please note: This video process is for cases you need to do a full recognition of the object in your app. Adjust the steps based on which part of the object you wish to track.

For more information on how to record a video as input for Objects Targets you could also refer to our documentation section titled 'Object Targets: Create and Manage'. You can also refer to this Youtube video regarding "How to: Object Recognition video recording for Wikitude SDK" and to this how to tutorial regarding "Tutorial: Step by Step Guide for Creating an Object Target Collection".


3. Having gone through all the bad examples for recording an object,  let’s have a look at a good example. To begin with, this object is a perfect candidate since the texture is ideal with edges and corners. In addition, the size of the object is not that small. Finally, the record process creates the perfect Object Target.



2 people like this