Method and system for automatic steering of an agricultural vehicle

0001The invention relates to a method for automatically steering an agricultural machine on a field along a surface to be machined, and to a corresponding steering system.

0002In recent years, many of functions were agricultural vehicles, the previously manually by an Operators were controlled, automated. In order for the operator to to facilitate work, for example, were different proposed steering systems which automatically on the vehicle is a direct field. You can facilitate the work of the operator and allow him to focus on other important parts of his work to concentrate, as a control of the utilization Working devices of the vehicle. It could also in the future be conceivable such vehicles without a driver on a field driving permit.

0003It automatic steering systems have been described based on different Principles based. Non-contact Systems can be divided in to a distance measurement based systems that attempt to the height difference between cut and uncut crop to detect, and in on an imaging-based systems that attempt to Differences in appearance between the crop to the cut and the uncut side of the crop material to detect or plant and the ground.

0004Examples based on a distance measurement systems can be found in EP 0887660 A and EP 1271139 A.

0005In EP 0887660 A is a to the bottom in front of a Harvester directed laser distance measurement device on an angular range pivoted. Based on their orientation and Position is the corresponding to the rotational angle position determines the point at which the laser beam reflected has been. In this way, a height profile of the soil determined which allows a swath or a crop material edge at the boundary between a harvested part of a to recognize cornfield and the as yet unharvested part and to generate a steering signal.

0006EP 1271139 A proposes with such a laser distance measurement device in addition before, the basis of the intensity reflected radiation, the amount of the male to determine crop.

0007A disadvantage of based on a distance measurement Systems is that a relatively expensive range finder with a laser and an electronic system for Runtime detection is required.

0008Examples of an image-based systems, with one targeted at the box camera and an electronic Image processing work, are described in DE 35 07 570 A, EP 0 801 885 A, WO 98/46065 A, WO 96/17279 A, ​​JP 01319878 A, JP 03 012 713 A, US 6,278,918 B, US 6,285,930 B, US 6,385,515 B and US 6 to find 490 539 B.

0009The DE 35 07 570 discloses an automatic steering system for a agricultural vehicle, the one along one or more than to go box with built in rows plants. The system directs the orientation of the rows of plants from an initial Down. Then the image along lines is scanned, extending parallel to the secondary orientation and the mean gray values ​​of each line are stored in a memory stored. While driving across the field, the field is along same lines scanned. The results of the scanning of each line are compared with the values ​​in memory, and be appropriate steering signals the reference values maintained.

0010In EP 0801885 A is a harvesting machine with a camera described, digitized the image signal and image processing is subjected. The rows are each on the Position a step function, at which the gray value gradually changes toward investigated. In each row the pixel positions with the highest probability intended for the step function. These positions be used to generate a steering signal.

0011In WO 98/46065 A also is based on the individual in the Lines detected information contained whether a Erntegutgrenze contain. will use the information of all rows found in which line the field being harvested ends.

0012WO 96/17279 A proposes a limit (z. B. Parallelogram) over a captured by a camera image to create and measure inherent in objects with a brightness that is above a threshold, to identify. The objects a regression analysis are subjected to their displacement to to determine a number of previously identified, which in turn to Generating a steering signal is used.

0013The JP 01319878 A relates to a steering system, in which Faces specific color in a means of a camera captured image will be connected to a straight defining line. This line is a row of plants viewed and used for generating the steering signal.

0014In JP 03012713 A is an automatic steering system for proposed a tractor, in which a window of a captured image is defined. This window is a virtual line defined on which are arranged the plants. The image with a virtual plant row is compared to generate a steering signal.

0015The US 6,278,918 B proposes the captured image into two or to divide four areas. In the areas will be a place and a direction of a row of plants intended. The better recognizable plant row is used to generate the steering signal used. The decision whether a pixel is a plant reproduces or not, is carried out by so-called K-means clustering, in which a histogram of pixel values ​​is produced, in which the x-axis shows the gray scale levels and the y-axis represents the number of pixels with the respective gray values. The Pixels that represent likely the plants appear, Therefore, at the right side of the histogram. The algorithm Parts the histogram generated from the image of the camera in a predetermined number K of classes whose intervals each are equally spaced. The data are processed recursively, to find the average gray value for the class. The Boundaries between the classes are based on a predetermined metric or minimizing a cost function recursively postponed until the centers of classes for less move than a given tolerance. Thereafter, a Threshold for distinguishing between plants and the ground placed at the beginning of class K or class. The classification can also go through other clustering algorithms done as allocation methods, self-organizing maps and neural logic or fuzzy logic. The US 6,285,930 B, US 6,385,515 B and US 6,490,539 B also refer to this system.

0016In the described, based on an imaging systems is determined from the brightness and / or color of each discriminated pixels, whether they represent a plant or not. The identified plants are in some of these Systems connected by a virtual line to produce a steering signal is used. It is, however, in some cases, in which an automatic steering of an agricultural Machine is desirable, not simply or even almost impossible, the color or brightness based to detect whether a Pixels to a line part, to be followed or not. A typical example is a swath cut grass or a harvested cornfield lying straw. The swath consists of the same material as in the field are remaining stubble and therefore in many cases more or less the same brightness and color. The described imaging systems are difficult to use here. Other Systems described recognize plants in a field, and based on the series described by them to deploy example chemicals. If no plants are present, these systems can not be used.

0017In the introductory part of EP 0801885 A (page 3, first paragraph) mentions a possible procedure under which a local two-dimensional Fourier transform operator as Base for a texture based segmentation used to a spatial frequency band with a significant difference between plants and cut standing plants determine. It is stated that this method at initial tests, no clear indication of such showed difference. It has therefore been rejected.

0018The problem of the invention underlying is seen in a method for automatically routing a agricultural machine and a corresponding steering system provide that works reliably and inexpensively can be realized.

0019This problem is achieved according to the teaching of claim 1 solved, in the other claims Features we have that the solution advantageously develop.

0020Disclosed is a method and apparatus for automatically Steering an agricultural machine along to proposed machined surface on a field. The machine is provided with a camera, the front of the field of directed machine including the area to be machined is. The camera provides at certain time intervals a two-dimensional Images of the surface to be processed and the next located areas of the field. From the image signal from the camera generated a bitmap file, the means of an electronic Image processing is further processed to a steering signal to produce and the machine automatically along to to perform processing area. An advantage of using a imaging system is opposite to a distance measurement based systems in the lower cost.

0021The invention proposes, from the bitmap file a texture information in texture around the individual derive evaluated pixel. There is an information generated which contains information about how the Structure of the image in the neighborhood of the pixel looks d. h. whether it is, for example, equally or unequally or the direction in which there existing details such. B. Ernteguthalme or clods, are oriented. look the Camera, for example on a mowed field with a swath, shows the field is substantially short vertical stalks and Swath longer straws with different orientations. the surface to be machined is analog in tillage different from the adjacent field areas, they as more or less, for example, after plowing or cultivating large clods and a more uneven surface shows as before. This texture information is the basis of a Classification of whether the respective pixel to be processed of the Area to be assigned or not (binary decision). Since not only as in the prior art, the brightness of the pixel, but beyond this information classification be taken as a basis, it is on a relatively safe Basis. The position of the surface to be processed can thus be recognized without problems. Based on the known position of the the surface to be treated (or not and / or still no surface to be processed) representing pixel is a Steering signal which supplied to the steering means of the machine becomes.

0022The texture information may be affected by a omnidirectional or directional texture analysis be won. In a non-directional texture analysis be information regarding the neighbors of pixel in all directions considered. In a directional Texture analysis, only the neighbors in certain directions evaluated. The direction-dependent texture analysis is particularly useful if preferred in the image characteristics in certain directions occur. The directional or non-directional Texture analysis, for example by means of a gray value dependency analysis be performed.

0023Alternatively or additionally, the texture by means of a Multispectral analysis are examined. Here, a camera used, the at least two different wavelengths of light can detect. In the different wavelengths can be distinguish the texture in the vicinity of the pixel so that you can decide whether the individual pixels respectively represent machining surface or not, an additional Information can lay a basis. As a multispectral provided a color analysis, in particular a RGB-analysis will.

0024The decision as to whether the individual pixels each to represent machining surface or not, is a so-called classifier hit. A suitable classifier is a neural network. The classifier preferably before the start of work by means of a supplied thereto And an image information about areas of the image, the show surface to be machined, and areas that are not to a show machining surface, subjected to a learning process. Of the Learning process, in the production in the factory or in the carried out starting work on a field.

0025The steering system according to the invention is particularly suitable for provided leadership of the harvester along a swath Harvested crops, as it in good discrimination between the stalks Swath and allows on the field. Such harvesting equipment for example, forage harvester with a pick-up as Crop intake, self-propelled or by a Tractor drawn balers and tractors with Trailer. The Swath is detected and the harvester is to him guided along. But the steering system described may also Harvesters are used along a crop material edge be moved, such as combines, forage harvesters with Corn harvesting machines or mowers, because the steering system in the Location is the long-standing stalks with ears of corn on the Top of the stubble on the harvested field to differ. Here, the harvesting machine with the outer edge of the crop pick-up at the edge of the crop passed along.

0026It should be noted, is that the image produced by the camera and a Information about the amount or rate of the male Crop contains. This information may be made to the contours of The harvested crop and / or of its color can be derived. The Steering system can thus calculate the expected rate and a Laying automatic control of the vehicle speed based.

0027Another possible application of the invention Steering system is the example by means of a plow, a Cultivator, a harrow, a roller or the like performed. Tillage or seeding. As already mentioned above, the texture of the soil is different after processing of the previously existing texture. The steering system recognizes the Boundary between the already machined part of the field and the unmachined part by the different textures and forwards the agricultural machine at her in a optimum spacing along, so that the outer limit of their Working tool at the boundary between the part being machined the field and the unprocessed part is guided along.

0028In the drawings, two embodiments are described further below of the invention. It shows:<dl tsize="7"><dt>Fig. 1</dt><dd>a forage harvester with an inventive automatic steering system in side view and in schematic representation,</dd><dt>FIG. 2</dt><dd>is a schematic side view of a Tractor with an automatic inventive solid steering system and one of him Baler,</dd><dt>Fig. 3</dt><dd>a block diagram of the steering system,</dd><dt>Fig. 4</dt><dd>an example of a camera of the captured image,</dd><dt>Fig. 5</dt><dd>a flow chart according to which the processor of the Steering system works,</dd><dt>Fig. 6</dt><dd>a schematic of a for the decision whether a Pixels to the field or to the swath heard usable neural network,</dd><dt>Fig. 7</dt><dd>the operation of the neural network of Figure 6 in a learning phase,</dd><dt>Fig. 8</dt><dd>the operation of the neural network of Figure 6 in an analysis phase,</dd><dt>Fig. 9</dt><dd>a returning to the image of Figure 4 Result image of a direction-independent gray value analysis,</dd><dt>Fig. 10</dt><dd>a returning to the image of Figure 4 Result image of a direction-dependent gray value analysis,</dd><dt>Fig. 11</dt><dd>from deciding whether the pixel to a Windrow or part of the field, going forth Image, which dates back to the image of Figure 9, </dd><dt>Fig. 12</dt><dd>the image of Figure 11 after removal of all Surfaces that are smaller than 450 pixels,</dd><dt>Fig. 13</dt><dd>the remaining, largest area of ​​the image in Fig. 12, and</dd><dt>Fig. 14</dt><dd>the end result of the processing of the image from Fig. 13 with marked inertia axes.</dd></dl>

0029In the figure, 1 is a harvesting machine in the form of a self-propelled forage harvester 10 shown. The forage harvester 10 is built on a frame 12, the front of driven wheels 14 and steerable rear wheels 16 will be carried. The operation of the forage harvester 10 is from a driver's cab 18 from which a harvested crop 20 cost. By means of the harvested crop 20 picked up from the ground, eg. As grass or like is not illustrated feed rollers within a feeder housing on the front of the Forage harvester 10 are arranged, a chopper drum 22 supplied that chops it into small pieces and a Conveyor 24 gives up. The crop leaves the Forage 10 to an accompanying trailer over one about an approximately vertical axis and rotatable in the Tilt adjustable discharge chute 26. Between the Chopper drum 22 and the conveyor 24 extends a post-chopper reduction 28 through which the to promoting the conveyor arrangement 24 fed tangentially becomes.

0030The crop pickup 20 is in this embodiment, formed as a so-called pick-up. The harvested crop 20 is built on a frame 32 and based on both sides mounted support wheels 38, are secured via in each case a support 46 on the frame 32, on the Soil from. The object of the harvested crop 20 is on the ground of a field in a swath 48 take Undocked crop and the harvester 10 to supplying further processing. To this end, the Crop pick 20 during the harvesting operation with short distance to the ground over the field moved while raised it to transport on a road or on paths becomes. For the crop pickup 20 includes a Conveying device 36 in the form of a screw conveyor, which the crop picked up from the sides of the harvested crop 20 located to one in the middle, not illustrated discharge port supports, behind the feed rollers consequences. The Erntegutvorrichtung 20 also has a, as well as the conveyor device 36, driven in rotation pickup 34 on which is arranged below the conveying device 36 and with its conveying tines the good from the ground raises to it the Conveyor 36 to pass. In addition, a Down 40 in the form of over the susceptor 34 arranged on the frame plate 32 is fixed.

0031The forage harvester 10 is at the top of in the direction of travel leading side of the cab 18 with a camera 42 equipped. The camera lens 42 is obliquely forward and down on top of the swath 48th The camera 42 is located on the longitudinal center axis of the forage harvester 10. The camera 42 forms with an electronic control, a below described automatic steering system that the forage harvester 10 automatically along the swath 48 to the work of simplifying operator in the cab 18th

0032First, however, with reference to FIG 2, another possible application the automatic steering system described. Here is the camera 42 at the top of the leading end in the direction of travel Side of the driver's cab of a tractor 18 50 appropriate. It is located on the longitudinal center plane of the Tractor and its lens is also forward and downward directed to the swath 48th The tractor 50 has front steerable wheels and rear driven wheels 54. He pulling behind a baler 56 ago, the crop, the out of the swath 48 by means of a transducer 34 from the field receives and formed into bales 58th Instead of the illustrated Round baler, the baler 56 a square baler be. You could also run self-propelled be. The invention may also to a soil-working serving tractor 50 used.

0033The forage harvester 10 and the tractor 50 are at their Front sides equipped with lighting devices, the allow to illuminate the swath 48, if the insufficient illumination.

0034In the figure 3, the steering system 60 of the forage harvester 10 is from Figure 1 or the tractor 50 of Figure 2 is shown schematically. The camera 42 has an image sensor 62 in particular in the form of a CCD or CMOS, whose output signal by means of an analog-to-digital converter 64 digitizes and Video interface 66 is supplied. The image data of the Camera 42 can only luminance (gray values, brightness) also contain chrominance (color) or exist. On Processor 68 (usually microprocessor or controllers) bidirectionally communicates with the video interface 66. a processor 68 connected to the display device 70 the can from the lens 82 of the camera 42 on the image sensor 62 generated image displayed to the operator in the cab 18 will. The display device 70 may also include other Information, such as operational data of the steering system 60 reproduce and / or of the forage harvester 10 or the tractor 50th The operator is also a keyboard 72 or other Input device (eg. A mouse, joystick, speech recognition) to Available, with which he perform inputs to the processor 68 can. The processor 68 controls electromagnetic valves 74, on the input side with a pressurized source 76 Hydraulic fluid and the output side to a steering cylinder 78 keep in touch. The steering cylinder 78 is adjusted in the Forage 10 from Figure 1, the rear wheels 16 and at the tractor 50 from Figure 2, the front wheels 52. A Wheel angle sensor 80 detects the current steering angle of the wheels 16 and 52 and supplies this information to the processor 68th The hardware configuration of such automatic steering systems 60 is known in the art sufficiently, so that there dispensed with a more detailed description and in the the introductory mentioned prior art Reference can be made, the disclosure by reference in the present document is received.

0035The task of the automatic steering system 60 is the forage harvester 10 or the tractor 50 with the baler 56 without steering action of the operator along the swath 48 to lead over the field. The processor 68 processes to the Images of the camera 42 by electronic image processing and generated based on the result of suitable control signals for the electromagnetic valves 74th

0036The swath 48 is usually made of dried stalks, the cut off with a mowing machine on the field to Drying left lying and merged by Rake or when harvesting a cornfield by the combine was deposited in the swath. A typical wherein Grass crop of the camera 42 captured image is in the figure 4 reproduced. It is apparent that the brightness values laterally between the swath 48 and the areas of the field 84 very little difference in addition to the swath 48th It would be thus problematic or impossible, a steering signal to produce by the individual pixels you (hereinafter: Pixels) of the image signal of the camera 42 based solely its brightness the swath 48 and the field 84 associates. On Chrominance data containing image could then better to Distinction between swath 48 and Section 84 on the basis of Colors be appropriate when the stalks in color sufficiently from the field is different. But this is particularly not the case, even when standing on the field remaining stalks are at least partially dried or if the swath is to be taken immediately after mowing, as in the grain harvest.

0037With reference to FIG 4 but can also be seen that on the field 84 essentially only vertically standing stalks (grass stubble) are present, while the swath 48 in contrast, extended and horizontally in different, certain to part extending orientations lying straws includes. The present invention therefore proposes that the texture of the pixels to evaluate and to decide whether a pixel on the Box 84 or the swath 48 is to consult. Among Texture is the local distribution and variation of the gray values (And / or chrominance values) in a portion of the image to understand. On the field, there is a texture of short, mostly vertical structures (stems), while in Swath of long straws is that in different Directions are present.

0038Accordingly, the processor 68 operates according to a procedure, as shown in FIG. 5 Starting from a the Step 100 captured image of the camera 42, in step 102 via the video interface 66 to the processor 68 a Pixel File provided. is from this pixel file in step 104, for each pixel processed further a texture information derived. In step 104, all of which can incoming pixel or only some of them, the one that Swath 48 containing part of the image included, are processed in order to shorten the calculation time.

0039The derivation of the texture information in step 104 may in different types occur. It is possible, a perform gray scale value dependency analysis, in which a is calculated as a function matrix. The Traceability Matrix contains small areas of the image information which Combinations of adjacent gray values ​​in a neighborhood the pixels occur. Are the gray values ​​in the neighborhood homogeneously of the pixel, thus, the dependency matrix thereon out that only identical gray values ​​are present. vary the Gray values ​​in the neighborhood, however very strong results, to another dependency matrix. In a preferred Embodiment, in the gray value analysis only direct eight neighbors of the pixel considered, at the edge of evaluated image area reduced their number to three (In the corners) or five. Obtained in this way Information as to whether the texture in the vicinity of the pixel is uniform or not.

0040The derivation of the texture information in step 104 can also by a direction-dependent gray value dependency analysis done. As can be seen from the figure 4 and already mentioned above, the field 84 contains a variety of vertical Line structures. In swath 48, however, the blades are wild confused. It is therefore advisable that diagonal to let neighbors of pixels taken into account and only the sub pixels and above, as well as left and right of the subject Pixel to create a dependency matrix to be used. The deduced in this manner in step 104, texture information thus contains information whether the texture in the Near the pixel contains vertical or horizontal structures. The amount of information, but not necessarily the information content, is compared with the non-directional gray value dependency analysis reduced.

0041Furthermore, the texture information in step 104 can also be a color analysis are generated. is in this study the chrominance information of the swath 48 analyzed. If a colored working camera uses 42, each pixel has the generated three color image with color information, each for red, Green and blue. If the swath 48 so color from the field 84 miscellaneous, can on the basis of color information easily between the two, even with the same brightness, distinction will. Finally, there is also the possibility of the gray level dependency analysis or the direction-dependent gray value dependency analysis to combine with the color analysis. It is therefore for each pixel evaluated omnidirectional or directional dependency matrix for the gray values ​​and generates a color information processed together will. In the said analyzes especially coming so said second order statistics are used. In addition, a serial combination possible in the first a pure RGB-analysis carried out, and then the result of the RGB-analysis a texture analysis is subjected.

0042After the step 104, information about the texture in the Around the pixel is obtained, according to the following figure 5 the Step 106. There is provided in step 104 on the basis of Texture information classification performed ie to be processed for each pixel or pixel group decided whether it belongs to the swath or field. Here can be any suitable algorithms. As Particularly advantageous is the use of a neural Network proved, since it is due to its ability to learn also can adapt altered image recording conditions. It is conceivable the use of a fuzzy logic or neuro-fuzzy logic. This decision is made in two stages: first is for each pixel or each pixel group a probability value evaluated, which represents the probability with which it or it belongs to the swath. It is then checked whether the respective probability value above a threshold value (in usually 50%) or not.

0043A possible embodiment of such a neural network 112 is shown schematically in FIG. 6 The network 112 is made up of at least two layers of neurons 114, 116 together. The first neuron layer 114 has an input 118, which obtained in step 104, texture information in the form fed the dependency matrix and possibly the color information becomes. In the neuron layer 114 a link which takes place Input values ​​with learnable information whose result to an output 120 of the first neuron layer 114 is ready and the input 122 of the second neuron layer 116 is supplied.

0044As shown, the first neuron layer 114 of second neuron layer 116 more output signals in parallel to the Available. The second neuron layer 116 has a single done exit 124. Also in the second neuron layer 116 a link to the present at its input 122 signals with learnable information. At the output of the second Neuron layer 116 is finally too binärisierende provided information on whether the examined pixels or pixel group for swath 48 or field 84 belongs. The Network 112 may also be further layers between the neurons two neuron layers 114, 116 shown comprise. It can also be any other network architecture than that of used shown, so-called back-propagation network will.

0045For each pixel to be examined could own neural Network 112 can be provided, or there is a single Network 112 is used, the input data for all the pixels are fed sequentially. In general, the neural Network 112 in software realized by the processor 68th In other embodiments, it could however also special hardware can be realized.

0046As mentioned above, the neural network 112 is capable of learning. He is thus initially taught which parts of a recorded Image for swath 48 and which belong to the field 84th This process is illustrated schematically in FIG. 7 the neural network (instead of from an image of the the camera 42 has been added, derived texture information) supplied the generated from a learning image texture information, that can be stored electronically, for instance, or the forage harvester 10 or the tractor 50 is still on the field 84 a swath 48 is positioned and the swath 48 is the Camera 42 added. Furthermore, the neural network 112 an information about the position of areas of the swath 48 and 84 of the field fed in the respective image. It can involve rectangles 86, 88, as in the figure 4 are shown. It masks are thus defined, in their Interior to the field 84 or the swath 48 belonging pixel volumes are included. The neural network 112 is based on this learning information in the location, as the swath 48 and the box 84 look like and to distinguish them. This Learning process could also be done by the driver to Forage harvester 10 or the tractor 50 by hand along a Windrow 48 controls. The neural network 112 learns also, between the swath 48 and the field to distinguish 84th The neural network 112 delivers a result image that the Success test serves. It can on the display device 70 be reproduced and inform the operator if the working conditions for the automatic steering system 60 are sufficiently or if, for example, in the dark or Fog, be better to resort to manual steering should. The learning phase of the neural network 112 may repeatedly repeated on a corresponding user specified way or lengthened, or differently than described above, carried out once during the manufacture of the steering system 60 or it can be used for changing conditions selectable Memory values ​​or removable memory cards provided will.

0047After the learning phase (Figure 7) is finished, the works neural network 112 in accordance with the figure 8. He is from Images from the camera 42 derived texture information supplied, and it provides a pixel image in which a binary done distinguishing swath and field. By subjecting the neural network 112 with the texture information is the distinction between the swath 48 and Section 84 relief and possible even at low visibility conditions.

0048Reference is 5 Referring back to the Figure in which according to step 106 now exists a binary bitmap file. The individual pixels are either the field 84 or the swath 48 assigned. Based on this bitmap is in the following step 108 the longitudinal axis of the swath 48 determines that it is their Direction (angle to the longitudinal central axis of the forage harvester 10, or Tractor 50) and their distance from the central longitudinal axis of the Forage harvester 10 or the tractor 50 determined. Based on the direction and the distance is then at step 110 a steering signal generated which is supplied to the electromagnetic valves 74th The forage harvester 10 or the tractor 50 moves thus automatically along the swath 48. The illustrated in Figure 5 procedure is repeated regularly, for example, 25 times in a Second. It is also higher or lower Repetition frequencies are used. At the end of the swath directs the operator to manually harvesting machine to the next male swath. In another embodiment, also the next swath through the automatic steering system 60 recognized and the harvester directed there automatically.

0049Figures 9 to 15 show illustrative images while that shown in Figure 5 from the processing in the Figure 4 Image shown emerge.

00509 shows a result for a non-directional Gray value dependency analysis. 10 shows a result by a direction-dependent gray value dependency analysis. In two figures are calculated by the neural network Probability values ​​for this illustrated that the pixel to Swath belongs before being binarized. A comparison of the Figures 9 and 10, both quasi the result of step 106 the figure 5 represent before the classification shows the positive influence of the direction-dependent gray value dependency analysis.

0051For further processing, therefore only the Results of directional gray value dependency analysis used. Figure 11 shows the result of decision (Step 106 in Figure 5) represents whether the pixel to windrow 48 (White) or field 84 (black) belong, ie the binarization by the neural network 112. The neural network 112 was previously subjected to a learning phase (Figure 7), in which it the image of Figure 4 as a learning image and the rectangles 86, 88 as to the field 84 or 48 belonging swath areas were notified. This network also showed other images good results.

0052Figures 12 to 14 correspond to steps of Step 108 of FIG 5. were all In the figure 12 contiguous areas calculated and all surfaces away, which are smaller than a threshold value (here 450 pixels). at Figure 13 is the largest area of ​​the figure 12 left been left in afterwards in Figure 14 by applying Dilation and erosion small black obstruct- were removed. In the figure, 14 also are the two Axes of inertia of the remaining area of ​​the swath 48 located. With the extending longitudinally Inertia axis, the direction of the swath 48 and its distance of the median longitudinal plane of the forage harvester 10 or the tractor 50 determinable. These variables are used to generate a steering signal used for the electromagnetic valves 74, wherein the current value of the wheel angle sensor 80 is taken into account.