Magic Leap reveals Deep Slam Tracking algorithm

July 28, 2017, 6:29 a.m. By: Pranjal Kumar

Deep Slam

Magic Leap may not have the headset to show but it is certainly making its mark in the technical world. The AR start-up has already reached a multi-billion-dollar valuation. In the recent paper published by Magic leap researchers, they have revealed the Deep slam algorithm. The aims at creating a robust standalone AR headset with the help of this algorithm.

The Deep slam algorithm describes a tracking system which is powered by the two-deep learning neural networks. The two-neural network is a type of an artificial brain which is used for image processing which is known as MagicPoint and MagicWrap. This allows the system to learn fast and can run at 30+fps on a single CPU.

Magicpoint operates on the single image and creates 2D points for tracking. These points are then fed into the simultaneous localization and mapping (SLAM) visual algorithm. This has proved to very crucial step as the result obtained are much better than the traditional way. As we know that calculation of shape of the moving objects can be a difficult task. MagicWrap used a pair of these images containing the 2D points (generated by MagicPoint) to predict the motion. This approach is a bit different from the traditional ways as it only uses the point’s location rather than using local point descriptors (a coded and unique information).

According to Magic Leap, the two-neural network is capable of running in real time. This can be a significant improvement in the traditional technology. Magic Leap has been associated with the rumor of developing the porotype of light field display packing head. The company calls the next headset as smaller, mobile, powerful and better looking than the present-day AR headsets. The recent development indicates that the company is making significant development toward achieving its goal.