Model-Free Marking in Augmented Reality

While geometric models are important in every Augmented Reality (AR) application, acquiring such model information is not an easy task, especially at long distances. We are proposing different 3D User Interfaces to help user mark a point's position in wide area both precisely and accurately.

Building on our successful work on Mixed Reality Simulation and AR simulation to enable informed design decisions, we present this novel AR project that can accomplish 3D marking task where the 3D position of a real world object is indicated by placing an virtual marker in the AR scene. To improve marking precision and accuracy at large distance, we explore different techniques, such as Geometric, Perceptual, VectorCloud and Image Refinement, that can help with the marking task when geometry of real-world environment is unknown. 

Geometric:

The user is asked to look directly at the marking target and indicate current head orientation. Then after moving with sufficient distance, the user needs to gaze at the same target and report the new gaze direction. Then the system can obtain target's 3D position via trangulation. The accuracy for estimation depends heavily on the distance between two oberservation locations.

Perceptual:

The user is asked to look directly at the marking target and indicate current head orientation. Then a virtual marker will appear in the AR scene immeidiately and the user needs to move the marker to desired distance along the direction. The accuracy is related to user's distance perception and can be improved by providing depth cues like relative size, linear perspective atmospheric attenuation.

VectorCloud:

This technique is a variation of Geometric technique using progressive refinement. As its name indicates, instead of only gathering two samples, it allows the user to send multiple direction samples. Since one single sample might not be accurate due to factors like user head trimmering, a refinement can be achived by using the average of multiple data samples. 

Image Refinement:

This technique intends to achieve high marking accuracy by decoupling direction sampling from head tremor. When the user starts marking, the system records the user's current field of view along with the camera pose and present as a static image. The user then pick the very point on the image to define a direction. The system calculates a precise ray in 3D space given user defined direction and camera pose information. Head tremor is completely avoided with this technique.

Collaborative Communication - Augmented Reality Rays:

This technique serves for model-free marking in with multiuser collaboration. When two distant users try to mark a common point simultaneously, communication is required to reach consensus on marking target. Virtual ray is a less effective tool to indicate either direction or select object when geometry information of the environment is not available. Thus we explore diffrerent visual enhancements to help separated users communicate and exchange spatial information.

Drone Asssited Marking:

This technique deploys a drone with onboard camera to take direction samples from a much larger space. By combining with Image Refinement marking technique, the drone's great mobility can help to tackle situaltions where the target location is either physically not approachable, or visually not available from ground.

 

Sponsors