Increasing Detail with Scanning


Fig 1. data can be captured and processed in under 1 second and points are independent, not averaged by block matching algorithms. The example illustrates one of the inherent advantages of Cloudburst’s low frame-to-frame noise. Data doesn’t have to be averaged, so additional frames are actually providing more and more detail. In the close-up of the box shown in figure 1 above, it’s easy to identify the shape and even depth of the folds and creases on the cardboard box. For reference, the gap here is approximately 2mm. The corner between the metal sheets is easy to detect and measure as well. This is the kind of detail necessary in many applications, and it’s easy to capture with Cloudburst. Scanning isn’t appropriate for every application, but when objects are in motion or the camera is mounted, for example on a robot end-effector, Cloudburst allows additional captures to increase the quality and detail in a 3D scene.

Fig 1. Close up of scanned pointcloud data showing detail on cardboard box (gap ~2mm)

3D cameras often demand a trade-off of resolution and detail for speed and flexibility. While expensive, multi-pattern 3D cameras are capable of capturing minute detail of a 3D scene, they are often very expensive and offer limited depth of field and field of view. Meanwhile, sensors based on stereo or time-of-flight can be inexpensive, and sometimes even hand-held, however detail is limited due to frame-to-frame noise, processing demands, or both.

Cloudburst offers a capable alternative. 3D data is collected with a single pattern of Symbolic Light™, and streamed at 15 frames per second. Multiple images from a scanned sensor can be merged, resulting in the kind of detail achieved by scanning laser systems, but in a fraction of the time.  The result is just like having 30,000+ scanning laser systems in a single camera.

Rock-steady 3D data adds up

Figure 2a shows the scene in which we capture a few objects, including simple planes, metal parts, small folds on a cardboard box, and a detailed poseable figure.

Figure 2b. shows pointcloud data from a single-frame capture of the scene. Cloudburst is factory calibrated, out of the box, so planes are flat regardless of distance or orientation. The rock-steady data makes it easy to find objects without averaging. However, even with tens of thousands of points in a single capture, some details, such as small gaps, or the fold lines on the box, can be lost.

2a. RGB photo of science

Fig 2a. RGB photo of science

Fig 2b. Single frame capture

Fig 2b. Single frame capture

Fig 2c. Scanned capture

Fig 2c. Scanned capture

 

 

 

 

 

 

By scanning Cloudburst at fixed intervals and capturing multiple frames, multiple pointclouds can be merged to fill in small details. Figure 2c is composed of several captures, where the camera is translated between each capture by 0.25 mm only along the x-axis; approximately parallel to the planes in the the background. Scanned data can be captured and processed in under 1 second and points are independent, not averaged by block matching algorithms. The example illustrates one of the inherent advantages of Cloudburst’s low frame-to-frame noise. Data doesn’t have to be averaged, so additional frames are actually providing more and more detail.

In the close-up of the box shown in figure 1 above, it’s easy to identify the shape and even depth of the folds and creases on the cardboard box. For reference, the gap here is approximately 2mm. The corner between the metal sheets is easy to detect and measure as well. This is the kind of detail necessary in many applications, and it’s easy to capture with Cloudburst.

Scanning isn’t appropriate for every application, but when objects are in motion or the camera is mounted, for example on a robot end-effector, Cloudburst allows additional captures to increase the quality and detail in a 3D scene.