SLAM in Augumented Reality | What is SLAM and importance of SLAM

Augmented reality is immature field yet to be explore more things but for exploring we need to be clear over basics, this post is about one of the basic concepts of the AR which SLAM , So lets explore how SLAM plays vital role in AR and makes its more realistic.

What is SLAM ?

SLAM refers to the process of simultaneously mapping moving objects with sensors, realizing self-localization in the process of motion, and modeling the surrounding environment in an appropriate way. SLAM think like be an optimization problem. For computing devices to understand, the device’s sensors collect visual data from the physical world in terms of reference points. These points help the machine distinguish between floors, walls and any barriers, and allows them to understand about physical space around there and makes decision based on that. The common SLAM system sensors are lidar, inertial sensors and cameras. SLAM with a camera as sensor is also called visual SLAM, which is VSLAM. Because the camera as a sensor collects abundant information, has strong object recognition functions, senses at much higher resolution than radar, has lower cost, is easy to carry, and has simpler operation, more and more people are paying attention to VSLAM and studying VSLAM technology. As the vision sensor may be a monocular camera, binocular camera or fisheye camera, VSLAM can be divided into monocular vision SLAM, binocular vision SLAM and RGB-D depth camera SLAM.

Figure demonstrate the working of VSLAM from high level prospective, which is also the classic SLAM framework. Different systems adopt different technical solutions in each module. VSLAM is mainly composed of four modules: Visual-inertial odometry (VIO), Optimization, Loop Closing and Mapping. Visual-inertial odometry (VIO) is the fusion of information measured by the visual sensor and by the inertial measurement unit. The latter is used to calculate the motion relationship of the sensor between adjacent time frames, and so calculate the motion trajectory. Whereas optimization deals with noise in the SLAM process. Loop Closing primarily solves the problem of error accumulation. The platform carrying the sensor constantly decides whether the environment is the same as at the starting point or the same location as at some previous time, and then decides whether to update the environmental map. In this way it is able to detect and correct errors and avoid incorrect information being introduced into the map. Mapping is a description of an environment, which may be in the process of movement. The constructed map can be one of four types: landmark map, metric map, topological map or hybrid map. Landmark maps, also known as sparse maps, consist of a number of landmarks. The Kalman filter (RKF) process uses landmark maps. Metric maps, also known as dense maps, represent basic space, two dimensional maps consist of grids or small squares, and three-dimensional maps consist of voxels or small cubes used for navigation. Topological maps are a more compact version of a metric map. They abstract a map as combinations of "point" and "edge", thus describing the connectivity between map elements. Mixed maps synthesize the advantages of various maps, avoiding the difficulties a single map may have addressing complex tasks. The specific manifestation of mixed maps is largely determined by sensor type, environment type and particular tasks.

SLAM Role

It is easy to navigate spaces that are known. But what about unknown pathways, “A known devil is better than an unknown one.

SLAM localizes and maps an unknown environment and navigates through spaces for which no prior map or GPS signal is available. SLAM is best applicable for situations with no prior reference point. SLAM is the amazing tech behind Google’s driverless cars. The autonomous technology that runs these self-driving cars uses a roof-mounted LIDAR sensor to create a 3D map of its surroundings. It does so within 10 seconds – quite a feat! The quick response is imperative in this technology since the machine in concern is a moving one with high speeds and acceleration. These mappings are augmented over the already existing Google maps. Through these readings, the autonomous system makes driving decisions using statistical algorithms like Bayesian filters & Monte Carlo models.

End Note

From the definition and need of SLAM we observed importance of recognition of physical environment for augmented reality, and with analyzing physical space around where objects are placed , another important factor is view setting for AR objects , here view represents the positions of the camera placements while recording or creating an 3D scene or object for particular use case, let’s assume an advertising company wants to place an AR holding over the public place in such a way that already existing view of holdings of same brand also be utilized and 3D representation put in way to enhance the way of representation of product in branding, in this scenario scenes are need to be created based on the already existed constrain which makes positioning and view as important factor. There are numerous cases of these types in real world which yet to be explore. So, with this conclude this post in which we explored about SLAM and need of it.

Thank you, readers happy learning!!!

Reference:”Visual SLAM: What are the Current Trends and

What to Expect?

Ali Tourani∗

, Hriday Bavle†

, Jose-Luis Sanchez-Lopez‡

, and Holger Voos§”, “An Overview on Visual SLAM: From Tradition to Semantic

Weifeng Chen 1,2 , Guangtao Shang 2

, Aihong Ji 3

, Chengjun Zhou 2

, Xiyang Wang 2

, Chonghui Xu 2

,

Zhenxiong Li 2 and Kai Hu 2,*”.