| With the development trend of intelligence and networking of automotive and the rapid iteration of centralized electronic and electrical architectures,ICV multi-source heterogeneous sensing components plug-and-play has gradually become an important research direction in the field of automotive electronics and intelligent driving.Since the implementation of 2017 National Key Research and Development Plan(2017YFB0102503),the plug-and-play platform architecture that realizes automatic networking,multi-mode decision fusion of heterogeneous sensing components and ultimately provides application layer plug-and-play,and safe,convenient,personalized driving experience is becoming a research hotspot.As the carrier platform of the next generation "Mobile Smart Life Ecosystem",ICV needs to have predictable environmental perception and positioning,as well as component customization and failure update functions while ensuring driving safety.Aiming at the lack of heterogeneous perception component platform architecture,low networking efficiency,and insufficient integration accuracy,the Pug-and-Pay platform architecture oriented to the perception layer has been studied and two key technologies of component automatic networking and decision fusion based on the architecture has been proposed.In addition,effectiveness and accuracy were verified by setting typical scenarios.The main research contents are as follows:(1)A multi-level and multi-dimensional plug-and-play platform and application division have been constructed for the requirements of the intelligent driving domain’s compatibility,difference,and convenience,as well as the platform’s hierarchical control safety and reliability.In the architecture,the physical layer,perception layer,and network layer provide computing power,storage services,environmental information,and data transmission resources for the application layer and make it realize plug-and-play in function,service and communication.(2)Aiming at the shortcomings and deficiencies of architecture functional layer networking,such as low search efficiency,poor calibration accuracy,high manual maintenance costs,a network process based on "Automatic Search-Autonomous Recognition-Active Calibration" has been proposed.This process strategy is through the process of network node graphs generation,node information transmission,characteristic parameters identification,spatial error compensation,realizing the plug-and-play of the application layer of heterogeneous perception components.The real-time,accuracy and efficiency of the proposed process were verified by PreScan-Simulink joint simulation platform,and the networking efficiency was increased by 30-40%.(3)Aiming at the integration requirements of heterogeneous sensing components after networking,a decision level fusion framework based on improved multi view K-means clustering and dynamic weighting function compensation was designed.First,the forward 3D target detection method was proposed innovatively based on the monocular Camera and the Lidar.Then,taking the target trajectory estimation as an example,a fusion framework considering time series was designed.Finally,the accuracy of the fusion framework was verified by PreScan-Simulink joint simulation platform,and the accuracy was improved by 7.6% and 11.8% respectively compared with the comparison algorithm.At the same time,the effectiveness and reliability of the above networking process were further confirmed.In summary,the performance of the perception layer in the intelligent driving domain from the three levels of "Platform architecture-Key technology-Experimental verification",such as high expansion,easy networking,and low coupling,were further broken and improved,and a more convenient and accurate plug-and-play implementation was formed,providing theoretical guidance and technical support for plug-and-play heterogeneous open unified embedded software platform design. |
PDF Full Text Request