Vehicle Position Measurement Method Based On Deep-learning Optimized Optical Flow Algorithm

In the modern war,the advantages of the Unmanned Combat Platforms(UCP)are increasingly valued by all countries.As the key technology of the UCP,navigation is regarded as the guarantee to guide UCP to complete various combat tasks such as tracking and positioning,investigation and early warning,and precision strike.Location information acquisition is an indispensable part of navigation technology.Optical flow algorithm is a commonly used velocity measurement method.In the process of obtaining UCP position information through integration,it is easily influenced by variation of illumination,relative movement of objects and accumulated error,which will result in inaccurate position information acquisition and poor-real performance.In order to further improve the accuracy and real-time performance of UCP position calculation,this paper carried out a series of optimizations based on the combination of deep learning and classic optical flow algorithm,and completed the feasibility verification through experiments.The specific process is as follows:Firstly,the classical optical flow algorithm is introduced,the advantages and disadvantages of several optical flow algorithms and their applicability are described,and a hardware implementation method of optical flow velocity measurement system is given.The Google Earth is employed as the simulation environment to evaluate the performance of pyramid LK optical flow algorithm in different environments,and 12-30 m is the most suitable height for optical flow velocity calculation.Secondly,the convolution neural network and clustering algorithm are employed to improve the pyramid LK optical flow algorithm,which reduces the influence of the relative movement of pedestrians,vehicles and other objects on the conventional optical flow velocity measurement method,and improves the accuracy of optical flow velocity.Experiment results indicate that velocity error of the improved optical flow algorithm is reduced from 4.66m/s to 1.58m/s.Then,the cubature Kalman filter based on multi-rate residual correction is used to complete the data differential fusion of optical flow and SIFT,compared with the pure optical flow system,the output frequency of the integrated system is increased by about 5 times,and the velocity error of the integrated system is reduced from 0.720m/s to 0.245m/s under illumination variation condition,which improves the real-time performance of the system and the immunity of light variation of optical flow.Finally,the position integration of optical flow is corrected by the position node correction mechanism,which is reduced the accumulated error caused by the integration.It can be clearly seen from the outdoor vehicle navigation experiment,the position error corrected by the position node is reduced from 11.5m / 500 m to 2.1m / 500 m,the feasibility and applicability of the proposed position measurement method are verified.