Font Size: a A A

Research On Ship Target Detection Algorithms For Maritime Video Surveillance

Posted on:2020-10-23Degree:MasterType:Thesis
Country:ChinaCandidate:Z H ChenFull Text:PDF
GTID:2392330602458474Subject:Traffic Information Engineering & Control
Abstract/Summary:PDF Full Text Request
Presently,in order to improve true positive rate and reduce false positive rate,the major impediment to maritime ship target detection is how fast and robustly detect the ship target under the various sea states that include a lot of sloshing water ripple,ship wake and illumination transformation.To overcome the above-mentioned obstacle,the thesis has accomplished the following research work.The maritime surveillance camera has various forms of motion during the shooting process,and there may be jitter or offset,which causes the ship target to be detected to be nonlinearly transformed and increases the difficulty of detection.In this thesis,the comer points with significant target features in the monitoring image were extracted,and the global.motion parameters of the current frame were obtained by taking the first frame as the reference and using the information between adjacent frames,and the motion parameters were filtered and compensated to eliminate jitter and output stable monitoring image sequence.In view of the fact that the monitoring equipment is installed on a non-fixed platform,there are jitters or offsets in the collected monitoring image sequence.At the same time,there are noise effects such as light transformation on the sky area and shaking water ripples on the sea surface.a novel method combining with the character of marine visible images is presented to detect ship targets automatically on the sea-sky background.First,do the preprocessing for the image,including video image stabilization and median filtering,Canny edge detection with surround suppression are applied to extract edge and eontour features for the visible images and one-place linear regression is used to pick the sea-sky-line and binarize targets above the sea-sky-line area.Second,in the sea-surface baekgroumd modeling,according to the feature of sea surface background,a k-means algorithm of oneself deciding the cluster number is proposed to detect the targets locating the sea surface area.Experimental results prove that target precision rate of the proposed algorithm is 81.1%,the false alarm probability is 7.3%,and the time cost per frames is 45ms,It is suitable for monitoring equipment installed on buoys or maritime Cruises on the ship target detection.The monitoring equipment is installed on a fixed platform.The monitoring background includes ship wake,illumination transformation and fish scale light.The monitoring device may jitter or offset,resulting in Monitoring image sequence shaking.a novel algorithm for ship target detection by using adaptive learning rate based on Gaussian mixture model(GMM)on sea surface is presented.First,do the video stabilization preprocessing to acquire video sequence after output stabilization.Second,the sky background,sea-surface and stationary objects are segmented quickly based on Discrete Cosine Transform(DCT)domain of image blocks.In the sea-surface background modeling,by dynamically adjusting the learning rate of standard GMM parameters,learning process can be divided into two stages include background formation and background maintenance updates,which can overcome the deficiency of background learning and excessive background learning.Finally,we implement detection results by using background subtraction and foreground segmentation.Experimental results prove that target precision rate of the proposed algorithm is 82%after the process of video image stabilization,which higher than that of without video image stabilization by 19%.The false alarm probability is 9%,which shorter than that of without video image stabilization by 37%.It can be used for monitoring equipment installed on port or wharf on the ship target detection.
Keywords/Search Tags:Traffic engineering, Video surveillance, Ship target detection, Gaussian mixture model, K-means clustering
PDF Full Text Request
Related items