Computer Vision Based Approach for Object Detection and Location Estimation in a Maritime Environment

Abstract

Maritime surveillance is a very important task in coastal areas, especially in harbour environments. The most popular such systems include components such as Automatic Identification System and Radar. Camera based visual surveillance can be used as an alternative to these systems in order to overcome the lacking features of them. Sea surface object detection and identification is a major need for such visual surveillance systems. Most of the current visual surveillance systems don t have the ability of identifying vessels in real time. A vessel can be identified using information from other systems, if the location of the vessel is identified. Location estimation of sea surface objects is mainly explored in this research. Video stream from a single geo stationary camera is used as the input, however camera properties are not used for any calculation. Location estimation using the proposed method can be used not only in surveillance systems but also in vessel traffic management systems. Tagging identified vessels in a real-time video is possible using this approach. It will be very useful when more information obtained from other systems are also displayed near the respective vessels. The distance of the object is calculated with reference to a known location on the image. Mainly two distance measurements are considered and different approaches for estimating the distances are explored. Interpolation approach for vertical distance estimation doesn t show good results while curve fitting and neural network approaches give considerably accurate results. Surface fitting and neural networks are used to estimate the distance to the object to the camera. Using a neural network give fairly better results than polynomial surface fitting. Best results can be obtained when B Spline 3D curve fitting approach is used. Data taken from Automatic Identification System is used for fitting curves and training the neural network. After calculating the distances, latitudes and longitudes are calculated. The calculated locations and the original locations are marked in a map so the accuracy of the calculated locations can be easily understood. An evaluation has been done comparing the calculated values and the values obtained from AIS data using various statistical tests. There, the different approaches are compared and accuracy levels are described.