Design And Implementation Of An Autonomous Driving Algorithm Test Platform Based On Environment Perception

In environment-aware algorithms for unstructured scenarios,there are high requirements for the multi-source datasets and the accuracy of the annotated information in the datasets.The existing datasets are mostly annotated manually,which is time-consuming and labor-intensive,but the annotation results are often not satisfactory.The rapid development of driverless demand and technology requires professional integrated test platforms,especially for comprehensive testing and evaluation of the functions and performance of environment-aware algorithms.Therefore,this thesis designs an environment-aware autonomous driving algorithm testing platform,which implements algorithm-assisted manual data labeling and algorithm cross-platform concurrent testing functions,and provides a platform labeled datasets and a highly concurrent algorithm testing environment for testing environment-aware algorithms.The main work of this thesis is as follows.(1)For the online annotation problem of environment-aware datasets,this thesis improves the existing environment-aware algorithm Faster RCNN in the feature extraction module,RPN network module and Fast RCNN module,respectively,and proposes Faster RCNN with difficulty prediction,so that it can be applied to the datasets collected by ground unmanned vehicles to assist manual annotation of the datasets.(2)For concurrent testing of environment-aware algorithms,this thesis proposes a hashbased distributed load balancing strategy,which is lightweight and non-intrusive compared to big data computing frameworks,does not require reconfiguration of environment-aware algorithms,and achieves an average time complexity of O(N)level for load balancing in terms of performance,enabling concurrent testing of different environment-aware algorithms and different multi-source datasets on the platform.(3)To address the implementation of the environment-aware algorithm testing platform,this thesis analyzes the requirements,designs and implements the environment-aware autonomous driving algorithm testing platform based on environment-awareness,and adopts the development mode of front-and back-end separation.The front-end uses Element UI component library of Vue.js to develop the interaction interface.The back-end uses Spring to integrate Spring MVC and Mybatis framework to provide data annotation and algorithm testing services.The database uses Min IO distributed file system to store unstructured environment-aware algorithm datasets.Finally,the platform designed in this thesis was developed and tested.The test results show that the environment-aware autonomous driving algorithm platform designed in this paper can meet the practical requirements of dataset storage and algorithm online testing,and reduce a large amount of labor cost,which has been tested in practice.