Design And Implementation Of Ship Borne Long-distance Target Positioning System Based On Binocular Vision

Intelligent unmanned rescue equipment such as search and rescue drone boats have effectively improved the ability to rescue people at sea and are expected to change the status quo of maritime rescue relying on heroes "trading lives for lives".However,the lack of effective positioning data is a bottleneck that hinders the application of unmanned boats,as people overboard are rarely equipped with positioning transmitters such as GPS,and radar has difficulty tracking small targets.Binocular stereo vision technology is used in many fields,such as medical and industrial applications,but there are few successful cases of target localisation on board.The reason for this is that traditional binocular vision localisation methods are unable to keep targets within the common field of view of both cameras while the ship is rocking.In this paper,a shipboard long baseline binocular vision target localisation method is proposed,in which two cameras with independent tracking functions are installed at any position of the search and rescue mother ship,and the internal and external parameters are calibrated at the initial position to establish a dynamic external parameter matrix with the gimbal angle as the variable;during the measurement process the two cameras acquire the target image separately,and the coordinate system is converted to the initial position by the rotation angle value to realise the spatial coordinate calculation.The thesis builds a free head-based long baseline binocular stereo vision maritime target positioning system hardware,and develops a friendly human-computer interaction interface,and carries out research to improve the applicability of target recognition,target tracking and binocular ranging in maritime application scenarios.In terms of target recognition,a background separation modelling algorithm incorporating colour(HSV)and texture(HOG)information has been developed to solve the problem of identifying weak targets on a wide sea surface;in terms of target tracking,the Camshift tracking algorithm has been improved by adding a pre-detector and calibrating the tracking results with N-expert and P-expert to improve the stability of target tracking under changing scenarios.In terms of binocular ranging,a dynamic external parameter matrix model is derived and its accuracy is verified by ranging.Finally,the system was tested in lake and sea trials.The results of the lake trial showed the accuracy of the range at 200 m line-of-sight,and the sea trial successfully guided the unmanned boat to approach the rescue target,which verified the feasibility of the system.