Ranging And Recognition Research Of Water Targets Based On Binocular Vision

At present,many researchers are researching on the intelligent and unmanned technology of water equipment.One of the key research directions is the environmental awareness technology of water equipment.As an important environmental perception technology,binocular stereovision technology has attracted the attention of researchers.In this paper,a series of key technologies in binocular ranging and recognition algorithms of underwater targets are studied.The feasibility and reliability of binocular ranging algorithm and target recognition algorithm are verified by experiments.The followings are the main researching content in this paper.Firstly,the technology of matching cost in binocular ranging is optimized and improved.An improved Census transform is proposed to solve the problem that the traditional Census transform relies too much on the central pixel and the matching effect is poor under the influence of noise.The improved Census transform is used for stereo matching.Experiments show that the robustness to noise and efficiency of stereo matching algorithm are improved.Secondly,the disparity extraction in binocular ranging is adjusted and optimized.Combining with goals of water target ranging and recognition,the contour of water target is extracted by Canny operator and Snake model with adaptive threshold,and then the adaptive window in the contour area is used when calculating the matching cost aggregation to avoid mismatching in regions that distance is discontinuous.Finally,the parallax is optimized to get the target parallax,and the distance information of the target is further calculated.Optimized and improved algorithm is used for ranging experiment.The ranging accuracy meets the requirements of obstacle avoidance and visual reconnaissance of unmanned aerial vehicle.Thirdly,target recognition and classification are carried out according to the characteristics of water targets.Because the texture of targets such as ships on water is obviously different from that of islands on water and has regular shape.So the shape feature,Hu moment feature and texture feature of the water target are extracted.The extracted target feature is used as the input of the classification support vector machine to recognize and judge whether the water target is a naval ship target or not.In the experimental verification,the support vector machine based on Gauss radial basis function is selected to train and test the samples,which achieves good classification and recognition results.