Research On Recognition And Location Method Of UUV Vision Based On Neural Network

Underwater unmanned vehicle (UUV) has been a kind of important tools of detecting seasince the human extended the development space to the ocean. When the UUV does the closerange operations, the location instrument such as sonar can not satisfy the positioningaccuracy requirements，so the light vision systems is widely applied in the UUV closeoperation tasks. In this paper，the problem of how to improve the recognition and locationaccuracy when the UUV doing the recovery work using Binocular vision location system hasbeen studied. The main research work is shown as follows：1. The pretreatment of Guidance image and the extraction method of Guide light source.The purpose of guidance image pretreatment process is to enhance the characteristics of lightsource, remove background and pseudo source. This paper uses gaussian smoothing filter toremove the noise. First the way of image binaryzation is used. Becauselight source andbackground gray feature is quite obvious, the immobile threshold can fulfill the request. Thenthe noise is filtered by utilizing morphological filter. In order to extract the light source profilefrom guidance image accurately，the binary image is deposed by the means of the Canny edgedetection，then the edge is extracted by to the Snake model extraction profile. The GVF filedis optimized by the greedy algorithm to improve the result and speed of Snake model. Regardto the feature extraction method, the Moment Invariants of light source is proposed as thefeatures to recognized target.2. The recognition method of heart of the reference light source. After extracting theMoment Invariants of light source，the Neural Network Algorithm is proposed as the classifierto solve the problem that how to recognize the light source is the heart-shaped or not. In orderto avoid the algorithm into local minimum，the BP network is improved by adaptiveadjustment，momentum factor and the method based on particle swarm optimized of BPnetwork. In the simulation，four kinds of BP network is carried on the training by the trainingsample. And then their speed are analysed and the best network is choosed for the next work.3. location system based on Binocular vision. When the UUV do the recovery work，theheading, transverse, longitudinal, height of the light source array should be ensured. It is easyto calculate the heading angle of the UUV with the matching line of the light source and thedirectionality of heart-shaped source. The method based on location system based onBinocular vision combined with Neural Network is proposed to calculate the freedom of the other three degree. This method avoids the error of the camera calibration. The experimentshows that the algorithms can achieve fast and effective with better usability, stability.4. Simulation of UUV’s the recovery work based on Binocular vision. First, theInstallation position of the guidance light source and the guidance system are designed.Second, the whole process of UUV recovery docking and Shrinkage scale simulation platformof the composition and function are Introduced. Finally, the software used in the simulation isdesigned based on the method which is mentioned above. The experiment which run in thesimulation platform with the data of UUV model shows that the method is available.