Research On Maritime Target Detection Method For Unmanned Surface Vehicle Based On Sea-sky-line

Unmanned surface vehicle(USV)is an emerging mobile carrier.The detection of maritime targets is crucial for USV to realize automatic obstacle avoidance and target reconnaissance.Compared to the radar image and the infrared image,the visible light image has more abundant information and cost of it is lower.However,the target detection method for the visible light maritime image is not yet mature.Considering that the sea-sky-line is an important reference for the detection of targets of the visible light maritime image,we carry out further research on sea-sky-line detection method.On this basis,the close-range target detection method and the long-range target detection method are studied respectively for close-range target avoidance mission and long-range target reconnaissance mission.Firstly,considering that the sea-sky-line detection is susceptible to wave edges,illumination and change of sea-sky-line gradient,a sea-sky-line detection method based on local Otsu segmentation and Hough transform is proposed.In order to extract sea-sky-line edges and suppress wave edges,an edge extraction method based on image segmentation is performed.For the precise image segmentation,the image is divided into image blocks to compensate for inhomogeneity of illumination,and then each image block is segmented respectively by using Otsu method.In order to suppress disturbing edges caused by error segmentation,the sea-skyline is fitted from edge pixels by using Hough transform.The results of test on visible light maritime image set show that the precision,robustness and timeliness of this method are better than three representative sea-sky-line detection methods.Secondly,in view of the missed detection and false-alarm caused by the diversity of closerange target and the complexity of background,a maritime close-range target detection method based on improved saliency computation is proposed.In order to suppress background and enhance target,the color of sky and the color of sea is estimated respectively,and then the saliency image is formed by calculating contrast between the color of each pixel and the color estimate of background.Pixels with a larger value in the saliency image are considered suspected target pixels,and suspected targets are identified according to the size and location of suspected targets area to suppress background interference.The results of test on visible light maritime image set show that target enhancement effect and background suppression effect of this method are better than three representative saliency computation methods.Furthermore,this method has high recall ratio and low false alarm rate.At last,on account of the missed detection and false-alarm caused by less pixels and low definition of close-range target,a maritime long-range target detection method based on edge scan filling is proposed.In order to extract target edges and suppress disturbing edges,the Canny algorithm is performed on the gray image,and then edges above and connected to the sea-sky-line are reserved.On this basis,pixels between target edges and the sea-sky-line are labeled by using the longitudinal scan filling method.The results of test on visible light maritime image set show that this method has high recall rate and low false alarm rate.