Research On Multi-Mode Positioning Data Fusion Method Based On Visual Optical Flow Vector Field

The global outbreak of the COVID-19 epidemic in 2019 has created a huge demand for intelligent robots with autonomous driving capabilities for outdoor epidemic inspection and food delivery.Such robots commonly use multiple Li DARs as positioning sensors,and the expensive price and huge arithmetic costs corresponding to massive radar point cloud data have limited their diffusion and use to some extent.A dual-mode positioning method combining a camera and an Inertial Measurement Unit(IMU),or integrating the Global Navigation Satellite System(GNSS)with an IMU,can solve the Li DAR positioning problem to a certain extent,but its accuracy is far from the actual application requirements.Therefore,this paper proposes a low-cost and less arithmetic-demanding multimode data fusion localization algorithm for visual optical flow/IMU/GNSS to improve localization accuracy,and conducts research in the following areas.Firstly,the paper analyses the navigation system structural components of an autonomous walking robot and the relationship between localisation,map construction and path planning.The data acquisition process and positioning principles of sensors such as Li DAR,visual optical flow,inertial navigation and GNSS are investigated,the spatial conversion problem and attitude solution process under different coordinate systems are analysed,the necessity of data fusion is discussed and the problems of existing data fusion methods are pointed out,and a framework of multimode positioning data fusion algorithm based on visual optical flow is constructed.Then,to address the problems of dense optical flow with high computational power and poor real-time performance,the sparse Lucas-Kanade optical flow algorithm is focused on matching and tracking feature points,and the Python-based Opencv machine vision development library under Win10 system is configured to write the sparse optical flow real-time tracking code.The fusion effects of the Extended Kalman Filter,Traceless Kalman Filter and Particle Filter algorithms are compared through simulation.The optical flow/IMU,GNSS/IMU fusion algorithm is designed under the Kalman filter-based prediction and measurement equations,and then the optical flow/IMU/GPS multimode positioning data fusion algorithm is completed.A solution to the temporal asynchrony problem of optical flow,IMU and GPS is given.Finally,a remotely movable multimode positioning data acquisition and processing platform with integrated IMU,GNSS and visual optical flow data acquisition functions was built and the sensors were calibrated.Multimode positioning data were collected outdoors,and trajectory maps of single and multimode fusion positioning data were generated to compare and analyse the positioning effects of multimode fusion and single mode.The applicable scenarios of the multimode fusion algorithm and the improvement methods are discussed.