Point Clouds and the Accuracy of Augmented Reality

Mobile augmented reality is by all accounts one of THE hottest topics to enter the mobile arena in some time. The idea behind augmented reality is relatively simple: overlay contextual or computer modeled information on top of real-time streaming information in context (and usually in 3D), and voila, youve got augmented reality.

There are plenty of examples in the consumer space where mobile augmented reality is being used for games, restaurant listings, and the like. If youre familiar with the popular consumer review site Yelp you may have tried their mobile (iOS) app, which comes complete with a feature called monocle. Monocle allows you to point your iPhone at a particular street and view the business listings (and average ratings) in context with the real view. Monocle, like most consumer mobile augmented reality apps uses the front camera, onboard GPS, as well as all of its sensors, like the digital compass and accelerometergyroscope (for positioning and angle), to present an augmented view of the scene in front of the device. While this produces a really cool effect for a consumer app, as you can imagine, given the accuracy (or lack thereof) of the onboard GPS and accompanying sensors, these kinds of mobile augmented reality applications are not really ready for primetime. When it comes to commercial application, like viewing underground utilities, or seeing a building concept in context before its built, the jitter effect caused by rudimentary GPS is simply not going to work.

Before I go on, its important to state that there are actually at least two kinds of augmented reality: one that uses markers, like visual survey monuments, and one that is marker-less, as in the examples above. As you can imagine, using markers to orient a scene and its digital overlay can produce very accurate results. However, this is also predicated on actually placing markers in the environment you want to augment, which also somewhat defeats the purpose in many cases. In other words, markers dont allow for much flexibility and it takes a lot of planning to place them. Marker-less augmented reality is the Holy Grail.

While onboard mobile GPS and sensors are certainly getting more accurate, something that may help with the accuracy of mobile augmented reality possibly enough to enable it to be used for engineering grade applications is the use of point clouds and imagery. By utilizing point clouds, maybe even coming from cheap devices, like the Microsoft Kinect, it’s possible to construct an accurate, real-time 3D model of the environment and at the same time track camera movement thereby enabling an application to snap digital contextual information or even concepts and designs onto the point cloud. In fact, even without the Kinect, it may very soon be possible to collect enough 3D information via mobile devices by way of dual cameras. Likewise, by using imagery (and planar information) to orient a mobile camera (along with all of the other sensors and GPS), its possible to provide much better accuracy. Companies, like metaio, are using many of these techniques today as part of their augmented reality development kits.

Another interesting approach is one taken by earthmine. Instead of relying on point clouds and imagery to augment reality, they actually use the entire 3D panoramic scene to do what they call augmenting reality. In this way, while the user may not be seeing the scene in real time, they may be seeing the best representation of the scene with engineering-grade accuracy to boot. Whats more, this kind of application could even be used at night, when the onboard cameras are close to useless.

As with other earth-based 3D applications, point clouds seem to be wiggling their way into the scene, mostly because they are accurate and 3D from the get go. And as point cloud creation techniques and devices continue to become more popular and drop in price, many people fully expect that point clouds will also open the door to new possibilities in digital media and design. While cheap GPS will become more accurate over time, you simply cannot deny that the real-time pinpoint accuracy that point clouds provide may play a big part in ultimately supporting engineering-grade accuracy for augmented reality.