Detection and positioning of the fruits is the first step in autonomous harvesting. Locating the fruit at outdoor light is a challenge for the vision system, especially the active systems. Passive vision systems, however, require extra matching algorithms which increase the detection errors. Judging whether a target fruit is pickable or not is highly depends on manipulator and environment, which is also under intensive research.
Bypassing the matching algorithm can reduce the error and thus increase the success rate of the passive vision system in the outdoor environment. A multi-camera passive vision system was developed without a matching algorithm. New neural networks and algorithms were implemented to judge whether a target apple is suitable for picking. For verification, a quad-camera vision system was built and tested. The functioning of the system was successfully proven.
Personnel - Ms Zijue Chen (PhD student)