Abstract：The intensity and depth information of reflected light were represented with image based on the analysis of TOF (time of flight) imaging technology. A new adaptive filter model coupling with partial differential equations filter and shock filter was established to remove noise and to sharpen edges efficiently. The algorithms of thresholding, region growing and smallest surrounding circle were used to segment mature citrus from the background. Depth range and variance within the selected region and radius of recognized citrus were calculated and used as constraint condition for further selecting of recognized citrus regions. Threedimensional database was integrated with extracted characteristic parameters of recognized citrus to realize realtime recognition and location. The total classification results show that recognition accuracy of about 86.7% can be obtained without false detection. The error of depth information is less than 12 mm with actual radius of citrus less than 13 mm. It takes less than 100 ms for image recognition and location.
2013, Vol. 34 
