^{*}

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/).

Logging harvesters represent a set of high-performance modern forestry machinery, which can finish a series of continuous operations such as felling, delimbing, peeling, bucking and so forth with human intervention. It is found by experiment that during the process of the alignment of the harvesting head to capture the trunk, the operator needs a lot of observation, judgment and repeated operations, which lead to the time and fuel losses. In order to improve the operation efficiency and reduce the operating costs, the point clouds for standing trees are collected with a low-cost 2D laser scanner. A cluster extracting algorithm and filtering algorithm are used to classify each trunk from the point cloud. On the assumption that every cross section of the target trunk is approximate a standard circle and combining the information of an Attitude and Heading Reference System, the radii and center locations of the trunks in the scanning range are calculated by the Fletcher-Reeves conjugate gradient algorithm. The method is validated through experiments in an aspen forest, and the optimized calculation time consumption is compared with the previous work of other researchers. Moreover, the implementation of the calculation result for automotive capturing trunks by the harvesting head during the logging operation is discussed in particular.

Because the circumstances of forest areas are very complex and hazardous, it is very dangerous and laborious to harvest standing trees by hand-operated machines and tools. In the last decade, with the development of the hydraulic controls and sensor techniques, more and more logging harvesters are being used in forestry [

The Forest and Environment Equipment Research Institute of Beijing Forestry University has been dedicated to the research on logging harvesters for years [

Laser scanner is a non-contact measurement system that can scan surroundings two or three dimensionally with a radial field of vision using infra-red laser beams. The distance between the laser scanner and the object is determined by the time of flight of laser light pulses: a pulsed laser beam is emitted and reflected if it meets an object. From those distance data, a point cloud is created describing the shapes of the objects surrounding the scanner. Laser scanning offers new possibilities in tree measurement applications in forestry.

Jutila, Kannas

In this paper, a low-cost 2D laser scanner and an inertial measurement system are mounted on the outer boom of the crane, and are used to obtain the point cloud of the surrounding trees. In Section 2, the whole laser measurement equipment is described, and a laser scanning experiment is carried out in an aspen forest; In Section 3, the hierarchical cluster algorithm and filtering algorithm are used to extract each trunk from the point cloud. The trunk radii and location of the trunks are calculated by the Fletcher-Reeves conjugate gradient algorithm; In Section 4, the measurement results are given and compared with previous work of other researchers; In Section 5, the implementation of the result for automated trunk capture is discussed. Our conclusions are presented in Section 6.

3D laser scanners are expensive and unsuitable for continuous measurements [

The 2D laser scanner can be installed on the outer boom of the crane, but the tilt and orientation angle of the scanner plane should be known. A MTi Attitude and Heading Reference System (AHRS) of Xsens Inc. is used. AHRS is a miniature inertial measurement unit with integrated 3D magnetometers, and is fixed on the top of LMS291 as shown in

The overall data analysis flow of the equipment for the measurement and calculation of the tree parameters is shown in

In our experiments, the laser scanner was fixed on a tripod with telescopic legs as seen in

In the experiment, the height of the scanning plane is equal to about 1.3 meters from the ground, this leads to better results because the understory and other uninteresting objects below the scanning plane and the variation in the height of the scanning plane is assumed to be negligible in our experiment. The measurement range is limited to 8 m, which is not beyond the reachable workspace of the crane of the logging harvester and can echo sufficient laser echo data from a single trunk.

In polar form, supposing vector _{i}_{i}, θ _{i}_{i}_{i}_{i}

In

The measurement data is processed in increasing order of the bearing angle from 40° to 140°. The cluster can be defined by two edge points, which are the measurement points that satisfy [_{i}^{th} measurement; Δ_{max} is the threshold for the allowed distance in distance inside a cluster. If the distance Δ_{max}, one of points ^{th} and ^{th} belongs to the cluster and the other one belongs to the background or other cluster. In our experiment, the distance between two trunks is large, so we choose Δ_{max} = 0.2 m, then, eight clusters can be extracted from the point cloud in

Using

The minimum and maximum curvatures of the whole cluster.

The minimum value for the curvature of a single point _{i}

The greatest acceptable width of the cluster.

The smallest acceptable depth of the cluster.

A point or cluster should be rejected if it fails any of the above tests. The minimum and maximum curvatures of the whole cluster in (1) prescribe the acceptable value of the trunk radius. The curvature of a single point _{i}_{i}_{i}, θ_{i}

After filtering and extracting the trunk clusters from the point cloud, the trunk clusters offer a variety of features that can be used for the logging operation such as radius, center location and distances between adjacent trunks.

Supposing that the matrix the _{1}, _{2},…,_{m}_{i}

Supposing the center location of the trunk is _{x}, O_{y}

Supposing the unknown vector _{1}, _{2}, _{3}]^{T} = [_{x}, O_{y}, R^{2} – (_{x}^{2} – (_{y}^{2}]^{T}, the coefficients vector

Then the scanning points in one cluster form a set of linear algebraic equations with constant coefficients as in the following

In order to get the parameters of the trunk, it is desired to solve for the unknown quantities _{1}, _{2}, _{3}]^{T}, given the coefficients _{ij}_{i}_{x}_{y}

^{m×3}, ^{3×1}, ^{m×1}, and:

For

Supposing _{k}_{k}

The vector _{k}

Therefore, the Fletcher-Reeves conjugate gradient algorithm (with restart) for solving

Step 1: Set initial condition _{0}.

Step 2: Compute _{k}_{k}_{=0} = _{0} = ^{T}_{0} – ^{T}

Step 3: Set _{0} = −_{0}.

Step 4: Compute _{k}_{+1} = _{k}_{k}_{k}

Step 5: Compute _{k}_{+1} = ^{T}_{k}_{+1} – ^{T}

Step 6: Compute _{k}_{+1} = −_{k}_{+1} + _{k}_{k}

Step 7: If _{max} go to step 4.

Step 8: Continue until convergence is achieved; termination criterion could be ‖_{k}_{max}.

_{max}is the maximum number of iterations. In our experiment,

_{max}is set to 1,000 and the initial condition

^{T}, consequently, faster convergence can be expected. The distance between two trunks in the horizontal plane can be calculated via the center locations of two trunks.

The trunk feature extracting process presented in the previous section is programmed with the Visual C++ 6.0 introduced by Microsoft and Matcom 4.5 introduced by MathWorks as MATLAB to a C++ compiler. All of the calculation results such as radii, location of the trunks and distances between adjacent trunks can be displayed on a human-computer interface for the operator.

The result of the calculations on the trunk parameter is given in

Extracting the parameters of the trunk from the point cloud also can be achieved by the Least Square Fitting algorithm (LSF algorithm) [

As shown in

It is found by the experiments that the process of delimbing, peeling, and bucking can be completed fast by the logging harvester, but for the process of the alignment of the harvesting head to capture the trunk, the operator needs a lot of observation, judgment and repeated operations, which lead to time and fuel losses. In order to improve the operation efficiency and reduce the operating costs during the logging harvesting operations, the laser scanner and inertial measurement system presented in this paper can be mounted on the outer boom of the crane to determine the location of the trunk near the harvesting head as seen in

In

Using the D-H parameters of the manipulator and combining the information of the harvesting head angle encoders, the pose and position matrix

Then, the pose and position matrix

The trajectory of the crane for trunk capture is planned in Cartesian space. The harvesting head moves smoothly one step in every control period until reaching the target. The motion planning and control flow for trunk capturing process can be summarized by the following steps:

Step 1: The laser scanner makes a fan-shaped scan of the surrounding area, and gets the location of every trunk in the scanning range.

Step 2: The pose and position matrix

Step 3: Using the

Step 4: Considering the constraints of joint velocity and acceleration, the central controller of the logging harvester plans the desired set points Δ_{k}

Step5: Using the inverse kinematics solution, the angle increments Δ_{1}.

Step 6: The angle increments Δ

Step 7: In the next ten control periods, the angle increments are acquired from the planned set points similarly, and hydraulic cylinders move according to the angle increments.

Step 8: if the capture range of the target trunk is reached, the gripper is closed, and finishes the mission, otherwise, it returns to step 1.

Because of the vibration of crane and control errors, only the first ten planned set points can be used for the task from step 4 to step 7, the rest of the planned set points should be discarded, and a new path planning calculation should be made in the next control period.

In this paper, in order to realize semi-automated logging harvesting, the point cloud for standing trees is collected with a 2D laser scanner. A cluster extracting algorithm and filtering algorithm is used to classify each trunk from the point cloud. The radii and positions of the trees are calculated by the Fletcher-Reeves conjugate gradient algorithm. Compared with previous work by other researchers, the calculation time consumption is reduced. The implementation of the calculation result on a logging harvester is discussed in particular.

The aim of this research relates to the human aspect of logging harvester operations. Logging harvesters are difficult to control and operator training is time consuming and expensive. If the location of every trunk relative to the harvesting head is known by the laser scanner, the operator could just indicate which tree to cut and the crane would automatically grasp it. The burden of the operator could be lightened by increasing the automation level of the logging harvester in the future.

This study is financially supported by National Natural Science Foundation of China (Grant No. 31070634), China Postdoctoral Science Foundation (Grant No. 2011M500009), 948 project supported by State Forestry Administration, China (Grant No. 2011-4-02), and Fundamental Research Funds for the Central Universities (Grant No. TD2010-2).

The whole hardware of measurement equipment.

The data analysis flow.

The experiment in the aspen forest.

The scanning and clustering result of the experiment.

The point cloud in clusters after filtering.

The calculation results.

Implement on the logging harvester.

The parameters of the trunks acquired by different methods.

_{x}_{y} |
||||||
---|---|---|---|---|---|---|

| ||||||

1 | (247.0, 248.0) | (251.4, 252.1) | (251.4, 252.1) | 8.6 | 9.9 | 9.9 |

2 | (397.9, 455.7) | (405.9, 465.8) | (405.9, 465.8) | 13.5 | 14.7 | 14.7 |

3 | (149.1, 520.0) | (152.6, 529.0) | (152.6, 529.0) | 13.6 | 15.2 | 15.2 |

4 | (−15.2, 782.9) | (−14.7, 785.9) | (−14.7, 785.8) | 13.3 | 15.6 | 15.5 |

5 | (−32.8, 272.0) | (−33.0, 276.0) | (−33.0, 276.0) | 9.0 | 9.3 | 9.3 |

6 | (−237.0, 597.7) | (−241.3, 608.5) | (−241.2, 608.2) | 14.6 | 17.3 | 17.0 |

7 | (−305.4, 346.7) | (−309.3, 349.6) | (−309.3, 349.6) | 10.3 | 10.4 | 10.4 |

8 | (−532.5, 471.1) | (−536.5, 476.3) | (−536.5, 476.3) | 13.9 | 15.2 | 15.2 |

The performance comparison of LSF algorithm and F-R algorithm.

Max error on the center location (cm) | (8.0, 10.8) | (8.0, 10.5) |

Max error on the radius (cm) | 2.7 | 2.4 |

Time consumption (ms) | 0.28 | 0.19 |