Our guidance technology exploits the known pattern in which a crop was
planted in order to distinguish between crop and weed plants in what can
be quite confusing scenes. Our main sensor, a CMOS or CCD camera,
is normally arranged looking forward and down at about 45º. We
view a number of crop rows so as to maximize the information available
within a scene. Colour images are transformed using a green/red ratio to
create grey level images in which green plant material appears bright
against the soil background with a degree of compensation to variable
natural lighting. Grey level images are filtered in such a way as to
pick out a planting pattern that has been pre-programmed into the
computer. This filtering process can be thought of as constructing a
template matched with crop plants visible in video images. It is
therefore necessary to enter the plant/row spacing into the computer.
This is normally achieved by selection from a number of pre-programmed
configurations. The more accurately the template fits the crop the
better it performs. Our systems display live video of crop rows
with template lines overlaid so users can check quality of fit at a
The astute will have noticed that weeds are also green, and probably will also be present if the machines objective is weed control! However, weed is normally randomly distributed and will not fit the template as well as the crop allowing the underlying pattern to show through. In general the more crop rows in view the stronger the pattern match and the better the immunity to weed or missing crop. If it is physically not possible to view as many rows as you would like with one camera then multiple cameras can be used.
The accuracy and reliability of dynamic plant tracking can be improved by tracking features from image to image taking into account knowledge of vehicle/implement kinematics and motion. For this reason it is useful to measure forward speed and side shift position/steering angle.Click on the image for a windows media video or on this link for a youtube version that will run on most platforms.
Having identified where crop plants are it is generally necessary to control some treatment device relative to those plants. In the case of inter-row cultivation or band spraying the control system is very simple requiring only lateral movement to align the camera, and therefore the toolframe attached to it, with estimated crop row position. This is normally achieved by a simple on/off hydraulic control system that operates a valve metering oil into either side of a hydraulic cylinder that shifts the toolframe relative to the tractor on which it is mounted. It is also possible to achieve the same result using a steered wheel to guide an implement that is connected to a tractor through a compliant linkage.Click on the image for a windows media video or on this link for a youtube version that will run on most platforms
Our innovative rotary cultivators that remove weeds between crop plants within-the-row requires the phase of rotary cultivation devices to be locked with the phase of plants as they disappear from the bottom of images. Our first commercial machines used fixed displacement hydraulic motors controlled by proportional flow control valves to maintain this phase relationship. We have subsequently developed brushless electrical drives that achieve the same result more efficiently and with more precision at up to 5 plants/s. Both these devises require sophisticated control systems with high speed control loops which are implemented on our own specially designed microcontroller based control boards.Click on the image for a windows media video or on this link for a youtube version that will run on most platforms.
Whilst this technology has to date been most commonly applied to mechanical weed control applications, an ability to track crop rows or individual crop plants also introduces other opportunities. These include thinning, spot application of material, chemical or biological, on to or around crop plants. It is also possible to store and georeference data for the purposes of mapping both weed and crop, though we have only done the later on an experimental basis.
We have recently completed further research aimed at the precision spot application of herbicides to individual weeds. In this case weeds are identified by a combination of factors including plant size, shape and position relative to crop rows. (see Projects)
In order to exploit a wide range of potential applications without introducing a large number of special "one off" solutions a modular approach has been taken to both software and electronic design. The multilingual user interface has a similar look and feel for all commercial applications and many machines can perform more than one function. For example it is possible for one machine to first mechanically thin a drilled crop, subsequently cultivate within-the-row and then as the crop grows too large for in-row cultivation to perform as an inter-row hoe.
Our main application software runs on standard industrial PC hardware which is packaged into a purpose designed cab mounted console. This provides machine control via a family of micro controller boards that we have custom designed to meet the needs of various applications. These micro controller boards are mounted on the implement and communicate between themselves via a CAN bus which facilitates connection in a variety of configurations. Typically machines use a master micro controller to control toolframe steering and levelling. To this hydraulic or electric in-row rotor control modules can be added along with nozzle control modules for spray applications. There are additional modules for mechanical feeler guidance and condition monitoring. (see Products)