Embedded vision detection of defective orange by fast adaptive lightness correction algorithm

Abstract

Automatic detection of defective fruit by computer vision system still faces challenge due to the uneven lightness distribution on the surface. A fast adaptive lightness correction algorithm implementation which is simpler and easier in real-time approach is proposed to overcome interferences from non-uniform reflectance intensity distribution on moving fruit surface and avoids error detection. The algorithm is tested by on-line and static defective orange images in the different lighting conditions. This study also compares other lightness correction algorithm implementations for defect detection. Recently, embedded vision systems are more and more popular because of low-cost, compact size and stability. A low-cost embedded vision system based on an industry gigabit ethernet camera and embedded Linux system in an arm processor with limited computing power compared with high-performance major PC is also originally developed to test and prove the performance of fast adaptive lightness correction algorithm using the most common surface defect of Navel orange in China. The time consumption of adaptive lightness correction algorithm of an image is below 6ms. The processing time of an orange image is below 30ms.