Research Of Background Modeling Based Fast Feature Matching Method And Its Application

In the study of the key problems of the multi-mode coupled visual and-haptic feedback in virtual surgery simulation,we have developed a visual-haptic interactive platform based on virtual reality,stereo vision and electromagnetics.The platform acquires the spatial position information of the operating instrument through stereo vision technology and transmits it to the virtual scene.The virtual scene utilizes the position to synchronize the virtual operating instrument and calculate the position between it and the virtual model,then haptic rendering is performed and the result will be transmitted to the electromagnetic feedback device.According to the size of the force,the force feedback device regulates the current,which causes the change of the magnetic field around the operating instrument,so as to realize the purpose of force feedback.During the positioning process,for avoiding the interference from the environment a white mask is used to cover the operating space,making the operation space and application scenarios of the platform limited;due to the occurrence of mismatching and a large amount of calculation of the traditional feature matching methods,the virtual operating instrument shakes in the virtual scene and real-time requirements can not be met;the usage of webcameras lowers the autonomy and integration level of this platform.Therefore,the existence of these problems is not conducive to the further popularization and application of the platform.In order to solve the problems aforementioned,this dissertation carrys out a research on the background modeling and fast feature matching based fast visual positioning method,and the main content of this work includes: study of background modeling in subtracting the non-target region,study of optimization methods for improving the quality of samples in statistical background models,study of fast and accurate feature matching method in calculating the space position,and the design of the hardware architecure and algorithms for fast positioning methods in embedded environment.The main research work is as follows.We propose a novel background subtraction framework based on kernerl density estimation(KDE).Firstly a new data structure called Mino Vector(MV)is designed for each pixel;we define dynamic nature(DN)for pixels of a scene and rank them in terms of DN for getting quantized results named dynamic rank(DR).Then,the varying KDE is adopted and implemented which significantly improves the estimation accuracy.Unlike using a global threshold in literature,we adaptively set a threshold for each pixel according to its DR.Inspired by the popular computer game Tetris,we present a Tetris update scheme(TUS)to update the background model in which the bottom row will be cleared,so do noises when the update condition is met.In experiments,we evaluate our framework on a well-known video dataset,CDnet2012.Our results indicate that our framework achieves competitive results when compared with the state-of-the-art methods.We propose an optimiaiton method for promoting the quality of the training data of background models.The quality of samples(QoS)for training has long been ignored.There are two aspects regarding this issue,which are how many samples are suitable and which samples are reliable.To tackle the ‘how many' problem,in this part,we propose a convergent method,coined Bi-Variance(BV),to decide an appropriate endpoint in the training sequence.In this way,samples in the range from the first frame to the endpoint can be used for model establishing rather than using all the samples.With respect to the ‘which' problem,we construct a pixel histogram for each pixel and subtract one from each bin(called NoIV-1),which can efficiently get rid of outliers.Further,our work is of plug and play that it could be applied to diverse sample-based background subtraction methods.In experiment,we integrate our scheme into several state-of-the-art methods and results show that the performance of these methods in three indicators,Recall,Precision,and F-Measure,has been improved from 4.95% to 16.47%,from 5.39% to 26.54%,and from 12.46% to 20.46%,respectively.We propose a novel constraint called approximately consistent in orientation(ACIO),which depicts the spatial location relationship of two matched vectors between stereo pairwise images(SPI),and therefore improves the accuracy of matching efficiently by avoiding the wrong correspondences.Secondly,we analyze the structure of standard K-d tree(SKD-tree)and propose a new one with hierarchical structure,named HKD-tree,which partitions the feature sets of SPI into stripes in terms of ACIO constraint and builds maps between them.By reducing the search space,the matching speed increases greatly.Thirdly we present an efficient and fast matching algorithm based on ACIO and HKD-tree.Extensive trials based on a benchmark data set show that the proposed approach outperforms the state-of-the-art methods in matching speed with slight promotion in accuracy.Particularly,it is one order of magnitude faster than SKD-tree and also several times against the recent CasHash method.In order to improve the integrity of the platform and perfect its independent intellectual property rights,we propose a two-layer hardware architecture design based on the Arm Cortex-M series high-performance microprocessor,and optimize the background model and feature matching algorithms.The front-end chip(a.k.a.the first layer)is used for signal acquisition and carrys out the optimized background model for filtering targets,while the back-end chip(a.k.a.the second layer)is used for sending the synchronous control signal for the image acquisition and matching features of filtered target objects,and then calculating the target's spatial position.The experimental results show that the design of the two-layer hardware architecture plus optimized background model and feature matching algorithms can realize fast position calculation in ensuring the accuracy,which lays a foundation for the subsequent development of our visual-haptic interactive platform towards high integration degree and wide application.