Agriculture Reference
In-Depth Information
TABLE 7.1
Performance of Fruit Detection
Total Number of Fruit
Detected
Rate
Positive detections
436
323
74%
Positive detections with clusters
436
415
95%
False detections
436
0
0
path optimization algorithms can be implemented here to minimize cycle time and
energy consumption. The robot is moved to orient the target fruit in the center of
the image by visual servoing, which improves the range estimation accuracy of the
ultrasonic sensor. Maintaining the fruit in the image center, the robot approaches
the target fruit. Once the robot is at a preset distance away from the target, the end
of the tool is moved a preset distance downward while continually tracking the tar-
get. This technique provides additional target range information through a stereo
vision technique. Once the range is estimated, further image processing is executed
to determine if the target is a single fruit or a cluster of fruits through a combination
of edge detection and circle detection. The within-cluster top fruit is selected as the
target, and the robot harvests the fruit. Then the robot returns to its start position to
search for a new target.
The fruit recognition algorithm was applied to a set of 24 randomly selected
images taken from the grove under different lighting conditions. Table 7.1 shows
a positive detection rate of 74% when ignoring fruit clusters, whereas a recognition
rate of 95% was achieved using the declustering algorithm. The inability to detect
the remaining 5% of oranges can be attributed to poor fruit color and occlusion. It is
also important to point out that there were no false detections in the tested images.
7.4.3.5 Fruit Picking Trials
The robot was tested in an orange grove with the robot positioned near the outer can-
opy of an orange tree using the macropositioning system as shown in Figure 7.32b.
A total of 450 harvesting attempts were performed. The robot had 357 (79.33%) suc-
cessful attempts and 93 (20.67%) failed attempts. Table 7.2 summarizes the causes
of the failed harvesting attempts. Most of the failed attempts were caused by range
TABLE 7.2
Sources of Harvesting Failures
Number of Failed
Attempts
Causes of Failure
1. Range estimation with ultrasonic and triangulation
31
2. Grabbing of multiple fruits
21
3. Failure to grab fruit due to occlusion
20
4. Inaccurate fruit center
18
5. Occlusion problem during approach
3
