^{1}

^{*}

^{2}

^{1}

^{1}

^{1}

^{1}

^{3}

^{3}

^{3}

^{1}

^{4}

^{4}

^{1}

^{1}

^{1}

This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

In this work, a LIDAR-based 3D Dynamic Measurement System is presented and evaluated for the geometric characterization of tree crops. Using this measurement system, trees were scanned from two opposing sides to obtain two three-dimensional point clouds. After registration of the point clouds, a simple and easily obtainable parameter is the number of impacts received by the scanned vegetation. The work in this study is based on the hypothesis of the existence of a linear relationship between the number of impacts of the LIDAR sensor laser beam on the vegetation and the tree leaf area. Tests performed under laboratory conditions using an ornamental tree and, subsequently, in a pear tree orchard demonstrate the correct operation of the measurement system presented in this paper. The results from both the laboratory and field tests confirm the initial hypothesis and the 3D Dynamic Measurement System is validated in field operation. This opens the door to new lines of research centred on the geometric characterization of tree crops in the field of agriculture and, more specifically, in precision fruit growing.

The geometric characterization of tree orchards is a non-destructive precision activity, which entails measuring and acquiring precise knowledge of the geometry and structure of the trees [

The measurement and structural characterization of plants can be carried out remotely using a number of detection approaches. These include image analysis techniques, stereoscopic photography, analysis of the light spectrum and ultrasonic sensors [

An Intelligent Laser Ranging and Imaging System (ILRIS-3D) sensor was used in [

A review was undertaken in [

Two papers [

In this work, a LIDAR-based 3D Dynamic Measurement System is presented and evaluated for the geometric characterization of tree crops. Using this measurement system, trees were scanned from two opposing sides to obtain two three-dimensional point clouds. After registration of the point clouds, a simple and easily obtainable parameter is the number of impacts received by the scanned vegetation. Given that their main function is photosynthesis, the distribution and position of leaves is clearly related to the availability of light. For this reason, the preferred position of leaves is normally in the outer part of the crown. With these premises, this work is based on the hypothesis that there may exist a linear relationship between the number of impacts of the LIDAR sensor laser beam and the tree leaf area.

The specific objectives that are considered in this work are as follows:

To evaluate of a 3D Dynamic Measurement System based on the 2D-TLS SICK LMS200 LIDAR sensor (SICK AG.) in dynamic conditions and at small laboratory scale.

To study the relationship between the number of impacts of the LIDAR laser beam on the vegetation and the leaf area of that vegetation. This study was first conducted under laboratory conditions using an ornamental tree and, subsequently, in a commercial pear tree orchard.

Section 2.1 describes the LIDAR sensor used in the laboratory and field tests. Section 2.2 details the specific materials and methods used in the laboratory work. Section 2.3 details the specific materials and methods of the field tests conducted in a pear tree orchard.

The terrestrial SICK LMS200 LIDAR sensor was chosen for this study (

The LMS200 is an eye safe (Class 1), time-of-flight laser sensor that emits at a wavelength of 905 nm (near infrared). Collaborative targets with specific reflectance features are not necessary and no lighting is required other than that provided by the emitted laser beam [

A total of 181 distance measurements using an angular resolution of 1°. These were obtained from a single complete rotation of the mirror (

A total of 361 distance measurements using an angular resolution of 0.5°. These were obtained from two complete rotations of the mirror. Obtaining measurements with 0.5° angular resolution requires twice the amount of time compared with a 1° angular resolution.

The number of measurements per second was the same with both angular resolutions. The RS-232 data transfer protocol was used between the computer and the sensor at a speed of 38,400 bits per second. It was verified that at this communication speed the sensor performs 1,700 distance measurements per second.

The direction of the Y-axis is a straight-line trajectory, which coincides with the forward motion of the LIDAR system. It is very important for the LIDAR system to travel at constant speed and follow an accurate straight-line path. In order to determine the value of the Y-coordinate of each single scan (slice), the developed software stores the time (in milliseconds) between slices. Since the LIDAR system is moving in a straight-line and at constant speed, the transformation of time into distance is direct.

It was decided to set up a dynamic system in the laboratory before testing the LIDAR system in the field in fruit tree orchards [

The motorised multi-purpose test rail was 7.54 m long and allowed constant speeds of up to 2.3 km/h. This rail had a mobile structure, which could move in both directions and was driven by an AC motor, which is controlled by a variable-frequency drive. The LMS200 sensor was mounted on this structure (

Since this study was carried out during the winter period an evergreen tree was used. A 2 m tall

The interior of this structure was sub-divided into rectangular prisms by using a nylon thread framework (

The minimum distance between the LIDAR and the front of the reference structure was 500 mm. The minimum distance between the LIDAR and the mid-plane of the tree was 1,000 mm. The reference structure was placed approximately half way along the mechanised rail. In this way there was plenty of time for the speed at which the sensor was moving to stabilise before passing in front of the

Angular resolution of the LMS200 sensor: The sensor was set to angular resolutions of 1° and 0.5°.

Travelling speed: The sensor was made to advance along the rail at 3 different speeds; 0.5, 1 and 1.5 km/h.

Orientation of the structure: Front and rear view scans of the

Using different combinations of these three variables, various LIDAR scans of the

All the leaves of the

The following procedure was used to determine the number of laser beam impacts in each of the 36 reference structure subdivisions:

LIDAR scanning of the

Visualisation of the three-dimensional point cloud using the AUTOCAD 2004 software (Autodesk, Inc.).

Visual localization and subsequent numerical determination of the coordinates of the reference point located at the base of the reference structure.

Running of the post-processing software to calculate the impacts in each subdivision.

After the winter time laboratory tests, the LIDAR measurement system was mounted onto a tractor and tested in the field to validate the initial hypothesis. These tests were conducted in a commercial pear tree orchard (

The two scans were subsequently registered into a single point cloud. To ensure the correct registration of the two scans, the tractor was displaced in a straight-line path at a constant speed of 1 km/h. It should be mentioned that the degree of accuracy in this respect was, logically, less than that of the laboratory tests. Four reference planes were also used, two on each side, to facilitate the correct registration of the scans (

In this section, the results obtained from the laboratory tests using an ornamental tree (

In the front view scans (1–6), the effect was studied of the angular resolution (1°, 0.5°) and speed (0.5, 1 and 1.5 km/h) on the distribution of the laser beam impacts on the test tree. It can be observed in

The number of impacts was approximately inversely proportional to the speed. The peaks of highest number of impacts for the six scans of different configuration appeared in the same divisions (boxes). This was also expected. The divisions with the lowest number of impacts were, in this case, the rear divisions (7, 8 and 9), since the elements in them (leaves and branches) were concealed by the leaves and branches of the front (1, 2 and 3) and intermediate divisions (4, 5 and 6) (

With respect to the number of impacts, it should be remembered that for each scan 181 distances are obtained with an angular resolution of 1° and 361 distances with an angular resolution of 0.5°. However, in the latter case twice the amount of time is required. Therefore, with an angular resolution of 0.5°, and at the same speed of advance, the vertical resolution (V) is doubled and the horizontal resolution (H) halved in the displacement direction of the LIDAR (Y axis). It is not the density of the points which changes, but their distribution. If

A verification was performed to compare the real dimensions of the reference structure with the dimensions obtained with the LIDAR system in all the scans. According to the data obtained with the LIDAR system, and after measuring the total height and width of the reference structure at various points, it was observed that the differences with respect to the real dimensions were ±1.5 cm.

After the

The front view F-05A-05S (Scan 1) and rear view R-05A-05S (Scan 7) scans were used to study the relationship between the number of impacts and the leaf area. Both scans were performed at a speed of 0.5 km/h and angular resolution of 0.5°. It was verified that, with this configuration, the resulting point mesh at the height of the sensor (beam angle of 90°) in the mid-plane of the

^{2}) values, 0.21 for the front view scan and 0.28 for the rear-view scan.

^{2} of the regression lines rise to values of 0.66 in the front view scan and 0.43 in the rear view scan.

^{2} of the regression lines (0.87 for the front view scan and 0.82 for the rear view scan), a good relationship can be observed between the number of impacts and the leaf surface area of the divisions closest to the sensor.

As we expected, the divisions furthest from the sensor received a lower number of impacts since they were concealed by the vegetation (leaves and branches) of the divisions closest to the sensor. For this reason, if we proceed to discard the obtained data starting from the rear and moving forwards towards the sensor, the relationship between the number of impacts and leaf surface area improves.

After separate observation of the front and rear view scans, the results were combined (^{2} of the regression line of all the points was 0.56 [^{2} of 0.89. This was obtained after discarding the “D” divisions, the lowest and furthest layer from the sensor, as their values can be considered outliers.

The relative position between sensor and vegetation determines the quality of the measurements of the vegetation. For example, if the sensor is situated close to and in the upper part of the vegetation, the lower part of the vegetation will logically be concealed or hidden by the rest and, consequently, the laser beam will be unable to reach this lower area. This is what happened in the lower layer “D” of the

It can be observed in

Having observed in the laboratory the good correlation between the number of impacts and the leaf area of a small

When designing this experiment special attention was given to the correct positioning of the LIDAR sensor. In order to guarantee the correct scanning of the vegetation, the sensor was positioned at an intermediate height of 2.1 m and an average distance from the vegetation of 2.5 m. This was done to avoid poor visualisation problems as occurred with the “D” layer of the

When the scans of the right and left hand sides of each section had been concluded, they were then registered (

It can be observed in ^{2} = 0.81) between the number of impacts and leaf area of all the divisions of the eight defoliated blocks. ^{2} = 0.87) between the number of impacts and leaf area of the eight defoliated blocks when ignoring the division into layers. By grouping together the divisions, the variability that arises from working with small units of vegetation is reduced and the correlation is significantly improved. If the “a” and “b” division between blocks in the 4 scanned sections is also ignored, R^{2} reaches a value of 0.89 [

The slope of the regression lines in the three cases is practically steady at 14.4. The vegetative state of the crop appears to have no influence on the relationship between the number of impacts and leaf area. It can also be observed in

The results obtained in the laboratory with the

^{2})/I, we have the equivalence between an impact (I) and the leaf area it represents (LA). Column 5, HV(cm^{2})/I, is the result of H × V. This value is the surface area of a grid of the point mesh, which represents or is equivalent to an impact. The results of 1.28 and 1.81 (LA/HV) for

The number and distribution of laser beam impacts on the tree used in the study (

The position of the sensor with respect to the vegetation is an important factor that needs to be taken into consideration as it affects considerably the viewing capabilities of the sensor and, consequently, the leaf area estimations.

Data capture from two opposing viewpoints considerably enhances the three-dimensional representation of the vegetation under study. Both the laboratory and field results confirm the initial hypothesis of the existence of a linear correlation between the number of laser beam impacts and the leaf area of the scanned vegetation. Replacement of the variable ‘number of impacts’ with the variable “impacted mesh surface area” is proposed to enable comparisons between tests using slightly different scanning meshes. This new measurement system opens the door to new lines of research related to the geometric characterization of tree crops in the field of precision agriculture.

As the LIDAR sensor obtains its measurements in polar form, one problem that needs to be analysed is the influence of the shape of the mesh in terms of the type of vegetation being scanned [

Analysis is also required in future studies of the equivalence between an impact and the surface area of a grid. Further studies should also be carried out on the influence of the size of the cross section of the laser beam and on the phenomenon of mixed pixels [

Given the observation that in small divisions (

In general terms, the continuation of the work undertaken in this study will involve specific studies using the LIDAR measurement system presented here. These studies will need to analyse a number of variables for different types of fruit tree of varying ages and different training and pruning methods.

This work has been funded by the Spanish Ministry of Science and Innovation and by the European Union through the FEDER funds and is part of the research projects Pulvexact (AGL2002-04260-C04-02), Optidosa (AGL2007-66093-C04-03) and Safespray (AGL2010-22304-C04-03).

LMS200 laser sensor (SICK AG) and its principal internal components.

Diameter of the laser beam cross-section of various models of the SICK LMS series and the separation between beams as a function of the angular resolution and distance to the sensor [

The impact points of the laser beam are determined in polar coordinates. The direction of the laser beam with an angle of 0° is vertical and upward pointing. The Cartesian coordinate system (X,Y,Z) used in the field and laboratory is also shown.

Front and top view of the internal division of the reference structure, together with the dimensions in mm and nomenclature.

Photograph of Section 1 pear tree vegetation (

Schematic description of the tests performed in a pear tree orchard (

Impacts of scans F-05A-05S, F-05A-10S and F-05A-15S with angular resolution of 0.5°, and respective speeds of 0.5, 1.0 and 1.5 km/h.

Impacts of scans F-10A-05S, F-10A-10S and F-10A-15S with angular resolution of 1°, and respective speeds of 0.5, 1.0 and 1.5 km/h.

Leaf area of each of the 36 internal divisions of the reference structure.

Scatter diagrams, regression lines and R^{2} of the relation between the number of impacts received in the 36 divisions (boxes) in the front view scan F-05A-05S

Scatter diagrams, regression lines and R^{2} of the relation between the number of impacts received in the 24 divisions (boxes) closest to the LIDAR sensor in the front view scan F-05A-05S

Scatter diagrams, regression lines and R^{2} of the relation between the number of impacts received in the 12 divisions (boxes) closest to the LIDAR sensor in the front view scan F-05A-05S

Left-hand side view of the reference structure and point cloud obtained from the rear view scan R-05A-05S

Scatter diagram, regression line and R^{2} of the relation between the number of impacts received in the 36 divisions (boxes), combining the front-view (F-05A-05S) and rear-view (R-05A-05S) scans, and the leaf surface area including layer “D”

Front view photograph of the

Point cloud obtained after registration of the scans of the left and right sides of Section 1.

Scatter diagram, regression line and R^{2} of the relation between the number of impacts and leaf area (cm^{2}) of all the divisions of the 8 defoliated blocks.

Scatter diagram, regression line and R^{2} of the relation between the number of impacts and leaf area (cm^{2})

Rear and front view scans of the

1 | Front | 0.5° | 0.494 | F-05A-05S |

2 | Front | 0.5° | 1.000 | F-05A-10S |

3 | Front | 0.5° | 1.520 | F-05A-15S |

4 | Front | 1.0° | 0.494 | F-10A -05S |

5 | Front | 1.0° | 1.000 | F-10A-10S |

6 | Front | 1.0° | 1.520 | F-10A-15S |

7 | Rear | 0.5° | 0.494 | R-05A-05S |

8 | Rear | 0.5° | 1.000 | R-05A-10S |

9 | Rear | 0.5° | 1.520 | R-05A-15S |

10 | Rear | 1.0° | 0.494 | R-10A-05S |

11 | Rear | 1.0° | 1.000 | R-10A-10S |

12 | Rear | 1.0° | 1.520 | R-10A-15S |

List of scans, performed on four different dates, of four sections of vegetation at a commercial pear tree orchard (

1 | 18-April | Left | S1 - (1a, 1b) | L1 |

2 | 18-April | Right | S1 - (1a, 1b) | R1 |

3 | 3-May | Left | S2 - (2a, 2b) | L2 |

4 | 3-May | Right | S2 - (2a, 2b) | R2 |

5 | 2-June | Left | S3 - (3a, 3b) | L3 |

6 | 2-June | Right | S3 - (3a, 3b) | R3 |

7 | 25- July | Left | S4 - (4a, 4b) | L4 |

8 | 25- July | Right | S4 - (4a, 4b) | R4 |

Summary of the results obtained in the laboratory tests (

^{2})/I |
^{2})/I |
||||
---|---|---|---|---|---|

3.34 | 2.9 | 0.9 | 2.61 | 1.28 | |

14.44 | 1.9 | 4.2 | 7.98 | 1.81 |