Rapid characterization of vegetation structure with a Microsoft Kinect sensor

Abstract

The importance of vegetation structure and biomass in controlling land-atmosphere exchange is widely recognized, but measurements of canopy structure are challenging, time consuming, and often rely on destructive methods. The Microsoft Kinect is an infrared sensor designed for video gaming that outputs synchronized color and depth images and that has the potential to allow rapid characterization of vegetation structure. We compared depth images from a Kinect sensor with manual measurements of plant structure and size for two species growing in a California grassland. The depth images agreed well with the horizontal and vertical measurements of plant size made manually. Similarly, the plant volumes calculated with a three-dimensional convex hulls approach was well related to plant biomass. The Kinect showed some limitations for ecological observation associated with a short measurement range and daytime light contamination. Nonetheless, the Kinect's light weight, fast acquisition time, low power requirement, and cost make it a promising tool for rapid field surveys of canopy structure, especially in small-statured vegetation.

Figure 1.

(a) The “tower mode” experimental setup used for nadir measurements with Kinect (see Sections 2.3 and 2.4). (b) Side view and (c) nadir view of a wild artichoke plant recorded during the field test (see Section 2.4). Blue and red lines represent plant's basal diameter and height as measured manually with a ruler.

Figure 2.

Comparison between nadir-only (a and c; collected using the “Tower Mode”) and co-registered multi-angular (b and d; collected using the “Multi-angular Mode”) point clouds. The upper panels show the point clouds viewed from an arbitrary horizontal perspective. The lower panels show convex (c) and concave (d) hulls superimposed as white lines. Panel (c) shows the shade (grey) and border (white) points used to define the convex hulls. Point cloud color scale is distance (meters) from the sensor at nadir.

Figure 3.

Comparison between raw infrared output from Kinect (a, c, e) and corresponding point clouds (b, d, f) in different light conditions. All images are for a potted rubber tree plant that was about 1.10 m tall (i.e., Figure 1(a)). Point cloud color scale is distance (meters) from the sensor at nadir.

Figure 4.

Direct comparison of manual and Kinect measurements of plant basal diameter (“Base”; the average of x and y measurements) and plant height (“Height”). Solid lines represent least squares linear fits.

Figure 5.

Direct comparison of volumes obtained from convex hulls and those derived from manual measurements with various solid shape approximations. Solid lines represent least squares linear fits.

Figure 6.

Allometric relations between Kinect-derived dimensions (base and height) and dry biomass measurements. Solid lines represent least squares logarithmic fits of the form y = A + B · log(x + C)/(log(D) + 1).