CurvSurf FindSurface™ GUIDemo for iOS (Swift)
This demo app demonstrates a real-time application using FindSurface to search point clouds, which ARKit provides, for geometry shapes.
FindSurfaceFramework is required to build the source code into a program. Download the framework here and refer to here for an instruction about how to setup your project to build it with the framework.
This demo app requires the LiDAR sensor to get a point cloud from ARKit.
- Apple iPad Pro LiDAR
- Apple iPhone 12 Pro
- Apple iPhone 12 Pro Max
- Apple iPhone 13 Pro
- Apple iPhone 13 Pro Max
Compared to the basic demo, the gui demo has the following features:
- Real-time detection of surfaces in camera scenes;
- Using input point cloud data provided by ARKit;
- Detected surfaces rendered in overlay (AR rendering);
- GUI interfaces to manipulate the app's settings (e.g., confidence, smoothed depth, feature types).
- Feature types of surfaces to detect.
- Hide/show the point cloud on the screen.
- Confidence levels to filter point cloud (e.g., the "
C.Med" option means filter out the points having "low" confidence). - Point cloud scanning mode.
- "Fixed(
Fix)" means that FindSurface takes about 20k points from ARKit as an input for each frame. - "Accumulate(
Acc)" means it takes about 500 points from ARKit as an input for each frame and accumulates them into a 500k points-sized buffer in a FIFO manner. The points are not accumulated unless the camera has a movement.
- "Fixed(
- Switch between smoothed scene depths and scene depths (ARKit option).
- Accuracy (error) levels of the target surface.
- Enable/disable FindSurface to start searching geometric shapes.
- Undo the latest snapshot.
- Remove all snapshots.
- Radius indicators.
- The seed radius indicator (the white circle), controlled by pinching the screen.
- The picking radius indicator (the red circle), controlled by dragging the screen vertically.
- The focusing point (the blue box) indicates the aim to the target object, to choose a seed point.
- (A) Target surface information.
- (B) Third-person perspective view mode (experimental, see the snapshot in the Output Examples)
- Points shown on the screen are the visualization of the points provided by ARKit, of which colors indicate confidence levels (green, blue, and red mean high, medium, and low, respectively).
FindSurface requires a priori root-mean-squared error of the measurement points (See the measurement accuracy description in this document for details). To determine an appropriate value of the measurement accuracy parameter, we built a linear model of normal noise for the app:
base_error(distance) = a + b * distance * 2
// the distance is in meters.
We set the values of a = 0.001 and b = 0.00005 for the normal (expected) noise model as a result of our experiments and introduced an additional constant values for each model, of which the values have been heuristically determined, to derive the following variations of the model:
lower_error(distance) = base_error(distance) - 0.001
normal_error(distance) = base_error(distance) + 0.001
higher_error(distance) = base_error(distance) + 0.003
The app calculates a concrete value of the a priori error according to the distance and the error level that users set (e.i., E.Low, E.Med, and E.High).
A sphere is detected
Points are hidden
Accumulated points
Third-person perspective view
