How to make a good trackable image for augmented reality

Image recognition is based on the number of key points that the platform can detect in the uploaded image (marker). The more points, the faster and more stable the tracking will be.

A good marker has the following features:

► High contrast. The difference between tones helps to detect more key points in an image.

· An example of a bad marker:

It’s hard to distinguish different details in the picture above because of a small range of tones. There are no bright highlights or dark shadows. Everything is in shades of grey.

· An example of a good marker:

This picture shows a wider range of tones. The object in the image stands out of the background.

► No repeating shapes or patterns. Identical or similar objects have a negative effect on recognition and tracking.

· An example of a bad marker with many identical shapes and patterns:

This image has many corner points, but they are repetitive and absolutely identical, therefore the tracking will be unstable as the points will get mixed up with one another. It is better not to use an image like this.

► Lots of details. The more unique elements in an image, the better it is recognised.

· An example of a good marker with lots of details:

► Even distribution of key points. Good image tracking requires that key points are evenly distributed throughout the image.

· An example of a bad point distribution:

Most of the points are gathered in one corner of the image, which will have a negative effect on tracking. There are some ways to avoid this without changing the marker. You can crop the image or add additional elements so that most of the points are not located in one corner of the marker, but are distributed throughout the whole image.

· An example of a good point distribution:

► Square and angular shapes. They have more key points than round objects. Round shapes are smooth without sharp transitions, so the image will be difficult to identify.

· Example of a bad shape:

Here we can see the shape without sharp lines. Such an image will be hard to detect.

· Even though a square has 4 corners, which are easy to detect... such a shape is not used for one reason — the square is symmetric, so it becomes impossible to determine its orientation in space. If you turn the square by 90º, nothing changes. That’s why it’s better not to use this figure as a marker.

· An example of a perfect shape is a triangle.

► The Feature — Exclusion Buffer If you look again at the images above, you will see that the points don’t fill the whole marker... About 8% of image stays unoccupied. (Even though there are objects in this area that can be detected). This area is an exclusion buffer. Point recognition doesn’t happen in this area.

You can avoid this situation by adding a white frame to the marker (about 8% of the width). The images below show an example of a bad marker and a good marker.

· Without a frame, precious information is lost. Some points are close to the edge of the marker.

· With a frame, the information stays within the borders of the marker.