Research On Crowdsourcing HD-Map Cloud Fusion Algorithm Based On Factor Graph Optimization

As one of the key technologies for autonomous vehicles,HD-Map can provide more accurate and richer map data than traditional navigation maps,and has a wide range of applications in autonomous vehicle positioning,perception and decision making.The traditional HD-Map mapping process is complicated and the cost of data collection is high,resulting in low update frequency of HD-Map.Crowdsourcing mode can greatly improve the data acquisition speed of HD-Map,reduce the acquisition cost,and obtain finer and more efficient map data,however,there is less research on the cloud fusion method of crowdsourcing HD-Map.To address the problems of crowdsourcing HD-Map cloud fusion methods,this thesis proposes a crowdsourcing HD-Map cloud fusion algorithm based on factor graph optimization,and the main work conducted is as follows:(1)A crowdsourcing HD-Map algorithm framework based on the " vehicle-edgecloud" architecture is designed.The framework consists of a perception layer,an edge layer and a central cloud layer,and the fusion of the crowdsourcing HD-Map is done by the edge layer and the central cloud layer.The serialization format of the lane map data is also designed according to the demand of cross-platform data transfer.(2)An absolute odometer generation algorithm based on GNSS data is designed to realize the globalization of local semantic lane maps reported from the vehicle.The Jacobi matrix of the constrained cost function of the optimization model of this algorithm is derived in detail to avoid the systematic errors brought by the iterative solver through numerical derivation.The accuracy of the lane map based on this absolute odometry is demonstrated by real data verification.(3)A crowdsourcing HD-Map cloud fusion algorithm based on factor graph optimization is proposed.The globalized lane map is used to realize the matching between map segments,a consistency screening process is introduced to improve the matching accuracy between map segments,and then a factor graph optimization model is constructed with lane line matching pairs to constrain the transformation relationship between maps.A detailed derivation of the Jacobi matrix for the constrained cost function of this optimization model is presented.(4)An evaluation criterion for the aggregation degree between lane maps is proposed.The experimental data are evaluated according to the evaluation criteria,and the results show that the crowdsourcing HD-Map cloud fusion algorithm based on factor graph optimization proposed in this thesis improves the aggregation degree between lane maps by 44.7%.Based on the collected reference truth data evaluated in the east and north directions,the results show that the error of the algorithm optimization results in this thesis is less than 1m in both directions,which verifies the practicality of the crowdsourcing HD-Map cloud fusion algorithm based on factor graph optimization proposed in this thesis.