Target Recognition And Location Of Underwater Vehicles Based On Vision Enhancement

With the development of outdoor field robots,the underwater robot technology has attracted more and more attention.Also,underwater robot has been gradually applied to practical applications.Underwater robots are becoming more and more important in the fields of underwater pipeline inspection,underwater aquaculture,ship physical examination,underwater entertainment,underwater archaeology and scientific research.At present,the operation of underwater robot is mainly done by artificial remote operation,which has the problem of low efficiency,time delay,lack of autonomy and intelligence,and so on.Improving the autonomous working capability of underwater robot becomes the primary problem and the key point of underwater robot research.This paper study the underwater image enhancement technology based on monocular vision in typical shallow pasture environment in depth.Based on this,the target detection and localization of autonomous aquatic products of observing underwater robot is realized.Specific research work is shown as follows:There are some shortcomings of low resolution,color slants and low lightness in underwater image.Thus,this seriously affects the ability of underwater robot.This paper studies on the visual enhancement problem of robot,and proposes the image enhancement algorithm based on improved Dark Channel Piror algorithm to preprocess on the monocular vision image.Firstly,the paper build degradation model of underwater image color slants and atomization phenomenon.Then,the image of depth imformation is calculated through the parallax of brightness channel and dark channel to estimate the background color of the water accurately.On this base,the adaptive scaling factor selection strategy is adopted to process transmission diagram for the higher contrast image restoration effect.In addition,the color correction method is taken to remove residual color slants and enhance the overall brightness of the image.The IDCP method proposed in this paper can obtain a better visual enhancement effect with better clarity and brightness than the conventional method under the premise of color correction.Different from the land environment,the underwater environment has the problems of low contrast,color bias and deformation caused by water flow.The deformation characteristics of the seafood itself exacerbate the difficulty of target detection and identification.In order to solve this problem,an improved Deformable Part Model detection method is proposed to detect and identify underwater targets in real time.Firstly,according to the distinctive color characteristics of the object in the scene,based on the HOG characteristics of the DPM algorithm,the color characteristics are integrated to improve the adaptability of the algorithm to the environment.Secondly,the deformable part model of the object is constructed,and the model is trained and learned offline using Latent Support Vector Machine.On the basis of obtaining the model,Fast Fourier Transform is used to convert the trained model to the frequency domain for convolution to improve the matching speed.At the same time,the fast construction that image features of the downsampling of the feature pyramid is used to improve the speed.At last,according to the score of various potential areas,the mapping model is used to predict the encirclement box.The algorithm can achieve a good balance between the accuracy and real time of target detection.Based on the detection and identification of underwater objects,in order to estimate the position of underwater objects,this paper uses the optical flow tracking method based on the SURF feature to track the eigenpoints and calculates the position and distance of the object by the means of proactive geometry.Firstly,the feature of SURF is extracted from the scene image,and the feature points in the highest score Mark box are selected by the detection algorithm.Secondly,the feature points in the mark box are tracked by optical flow method.If the feature points are lost,the SURF features are reextracted and tracked.Then,according to the position of the feature point and the pixel position of the tracking point,the robot's motion transformation is solved by RANSAC algorithm.Finally,the three-dimensional coordinate information of the tracking point is calculated by means of triangulation with the prior knowledge of the object,and the algoithm uses the average of the three-dimensional coordinates of these tracking points as the coordinates of the object.Under the condition of low cost sensor and robot body,the algorithm can locate the target object in real time,thus providing guarantee for robot underwater operation.On the basis of the above research,the underwater robot equipped with a monocular camera and other sensors is used as the experimental verification platform,and the proposed underwater image enhancement,object detection and location methods are tested in shallow water environment.The results prove the validity and feasibility of the proposed method.