Although this method allows for high-precision measuring, it has a drawback however. What it measures is the distance from the plant to the camera and not the height of the plant. Yet if the ground surface is unequal (which is always the case in fields!), the fact that the distance between the plant and the camera is the same for two plants (conditions for not destroying them) does not mean that the plants have the same height relative to the ground! Marie-France Destain continues, “it was therefore necessary to correct the distances from camera to object by modeling the land surface. We reconstructed the ground level which is under the plants by means of ground pixels visible on the images.” The system, which has been patented, has been integrated into a laboratory prototype and that has been very satisfactory, the recognition has been correct in more than 80% of cases. The rate is not 100% but, if it is implemented into a robot, the latter can move around continuously in the field in the same way as robotic lawn-mowers.

Acting during a specific period

There still remains a problem to be solved however: how will the robot know that it must intervene? Will it be necessary to introduce data on the height of the plants for example? Marie France Destain replies “No, in the beginning we considered that it was necessary to introduce the sowing date from which the system would determine the most probable height of the plants in accordance with this date. But there are climatic variations etc. Therefore we chose to work by taking account of statistical aspects related to the growth of plants.”



On the curves represented here, we see that during a certain period, carrots are really much bigger than the others. In this way we isolate a period during which the robot’s work will be optimal. In other words, we know that the system will not work if, for example, we intervene too late, as shown by the probability distribution curves, the difference in height between the plants and weeds will no longer be discriminating.  It is also useful to have a robot that passes through non-stop in order to intervene as early as possible. The system perfected at Gembloux was for the detection of plants in the planted row; outside the row use can be made of existing solutions, also mechanical, because robots for this kind of activity are already available on the market. The remaining research is currently aimed at the selective destruction system. The idea is to find the optimal system for destroying weeds bearing in mind that they are small: it could involve thermal destruction (burning by heat), or electric (a small discharge is sufficient) or mechanical (pulling).