With three different pictures of one plant, on some pictures the fruits were occluded while on another picture of the same plant, the fruit was totally visible. In this way more than 95% of the cucumbers were detected correctly. In a research on a cherry harvesting robot [7] different positions around the crop were studied to increase visibility. Images were taken from four different positions around the trunk of the plant. It was stated that 59% of the fruits were visible when all images were used. In research by [8] multiple positions around a citrus tree were investigated to determine the positions that were needed to get the highest fruit visibility. A combination of up to six views resulted in a significantly higher visibility.

In this research, the effect of multiple camera positions and viewing angles on fruit detectability in a sweet-pepper greenhouse crop was investigated. The objective was to determine the optimal camera position or combination of positions which yields the maximum visibility of the sweet-pepper fruits on the plant for the purpose of robotized harvesting. Little information can be found concerning the minimum visible fruit surface a computer vision system needs, to be able to detect and localize a fruit. According to [5] the segmentation for spherical objects (citrus) proves very efficient for spheres visible for more than 50% of their surface. To our own experience, fruits that are occluded for more than 50% are hard to localize with the precision required for robotic harvesting.

Therefore most results on fruit detectability presented in this research are under the precondition that at least 50% of the fruit surface is visible from a certain viewpoint.2.?Materials and Methods2.1. Recording DeviceA recording device was built which allowed placing the camera AV-951 under different azimuth and zenith angles. The device was placed on a crop handling cart so it could easily be moved along the crop row as well (Figure 1). A 1/1.8 inch CCD colour camera (Stingray F201C, Allied Vision Technology, Stadtroda, Germany) was used for the recording. The camera was equipped with a low distortion wide angle lens with a focal length of 4.16 mm (Lensagon CMFA0420ND, Lensation, Karlsruhe, Germany). On a distance of 0.5 m the field of view covered a crop area of about 0.7 m by 0.5 m. This area was sufficient to capture the region where all ripe fruits were located.

In the rare case a fruit on the plant was located outside the captured region it was not counted and not included in the analysis. The camera was mounted on a slide with a tilt unit which could be moved on a metal arc with a diameter of 1.0 m. Tilt unit and arc allowed setting the camera to different azimuth and zenith angles. Before acquiring images the recording device was positioned such that the recorded plant was located in the centre of the arc.