Studies And Applications On Inland River Ship Detection Technology Based On Video Sequence

Jiangsu Province is nearby the sea and full of lake and river.Due to the lack of coal and iron resources,bulk cargo relies on waterway.Waterway costs ten percentage to highway,while the safety precautions is far behind.We know that the highway in our province has taken full video surveillance,and technology of video analyzing is gradually used.When referring to technology from road traffic video surveillance,difficulties occur,such as unlike the road which has standard white lines,the water sparks and has no markings,which make it hard to calibrate the camera.What’s worse,ships’s different size and their stern ripple lead to considerable error when calculating ships’ width,length,speed,and flow.We work on the project "Research and Application of Key Technologies in Real-Time Video Monitoring and Recognizing System",and the studying of ships’ video surveillance has important theoretical value and great economic value.This paper describes the domestic and foreign status of ship parameters detection and design the Real-Time Video Monitoring and Recognizing System including system architecture,hardware,software modules.Basing on camera imaging model,it designs the interactive calibration algorithm.It compares several common methods of background modeling,and picks VIBE for ships’ contour extraction.To make the system running in real-time,GPU parallel programming is used.As to camera calibration,the paper designs an interactive camera calibration algorithm based on grid.It gets the conversion equation between image coordinates and the world coordinates by analyzing the pinhole camera imaging model,simplifies the process of calibration by reducing the number of parameters,and determines the camera calibration parameters through artificial observation of whether the generated rectangular grid is coincide with the ground.As to ships’ extraction,the paper analyzes the characteristics of common algorithms of moving object detection.To solve the problems exiting in object detection on water such as wave interference,weak contrast between target and background and stern ripple interference and so on,the paper picks VIBE and optimizes the process of initialization to avoid the "ghost" phenomenon,which may occur when the program starts.As to the situation that background modeling program takes too much time which makes it unable to run in real-time,the paper uses GPU technology to accelerate the algorithm,and the effect is obvious.Considering VIBE’s huge scale computation and that the pixels are processed independently,the paper analyzes the difference between CPU and GPU multi-core architecture,and adopted GPU architecture which is more appropriate for parallel programming,to accelerate VIBE.Experiments shows that the acceleration ratio under different resolution video is more than 10 averagely.The system is originated from national project "ship network system",and is applied on Nantong and has passed the national pre acceptance.At present,10 monitoring points is picked in South and North of Jiangsu waterway,forming the inland water video monitoring network system.The main innovations of this paper are as follows:● It designs an interactive camera calibration algorithm based on grid to solve the problem of lacking of reference object on water.● It optimizes the vibe algorithm,using multi frames to initialize the background model,eliminating the "ghost" phenomenon which may happen at the beginning of the program.● It uses GPU parallel programming technology to accelerate Vibe algorithm,making it possible to handle HD video in real-time.