Investigation of Image Processing Methods Capable of Supporting Navigational Lookout

摘要

This paper presents a new image processing method for extracting ships from marine images shot by monitoring cameras during navigation.Marine accidents are mostly caused by collisions between ships,the primary cause of which is improper lookout.Although the enhancement of lookout capability is important for preventing marine accidents,increasing the number of lookout standers is not easy.Therefore,the development of lookout support systems is necessary.The authors have previously proposed several ship extraction methods.Two of these methods and their disadvantages are described below:· To increase the accuracy of extraction,we developed a repetitive matching operation method.However,it required an enormous amount of processing time.· With the objective of increasing the speed of image processing,we employed the Lucas-Kanade Method (LKM).However,the LKM deals with pixel-by-pixel processing of the optical flow,thereby lowering the accuracy with which ships can be extracted.To utilize the advantages of both methods and reduce their shortcomings,this paper proposes a new method.First,the method divides an image into regions on the basis of the range of brightness.From the size of the divided regions,the regions corresponding to parts of the images of a ship are identified.Next,the optical flows are determined pixel-by-pixel using the LKM,and pixels are selected by calculating the inner products of the sequentially neighboring two optical flow vectors.Both the regions and the pixels selected are integrated on the image and the regions containing pixels with straightness are regarded as part of the image of a ship.By processing images shot by monitoring cameras on board,the effectiveness of our proposed method is examined.