| High-definition map is the key technology of autonomous driving,which has the characteristics of high precision,multi-dimensional and high real-time,and provides support for the perception,location,planning and control of autonomous vehicles.Because of the frequent changes of traffic roads,high-definition maps need to be updated repeatedly.The method of updating high-definition map by mapping vehicle is slow and high cost.However,the way of updating high-definition map by crowdsourcing expands the data collection range,accelerates the data collection speed and reduces the data collection cost.Therefore,the industry has paid close attention to the production mode of high-definition map with crowdsourcing.In addition,the development of edge computing technology solves the problem of data transmission and calculation in high-definition map crowdsourcing,and further promotes the development of high-definition map crowdsourcing.In this context,this paper proposes an edge computing based high-definition map crowdsourcing framework for autonomous driving,which can be divided into the revenue maximization oriented task publishing mechanism and the data quality driven region uniform crowdsourcing member selection mechanism.In the revenue maximization oriented task publishing mechanism,this paper finds that there are diminishing marginal utility and data aggregation effects in the data collection process of high-definition map crowdsourcing,and models the high-definition map crowdsourcing process and the problem of the crowdsourcing utility maximization.In order to meet the needs of crowdsourcing with high revenue and high time coverage,this paper proposes a highdefinition map crowdsourcing task publishing scheme,which adapts to the time division based on traffic flow,realizes the uniform and high coverage of crowdsourcing data in time,and controls the stop time of crowdsourcing task through the optimal stopping rule,so as to maximize the crowdsourcing utility.In the data quality driven region uniform crowdsourcing member selection mechanism,this paper first proposes a region partition mechanism based on perceptual features to achieve uniform coverage of crowdsourcing data in space.For the demand of high quality of crowdsourcing data,this paper analyzes the factors that may affect the data quality of participants,and models the problem of maximizing the quality of single time slice crowdsourcing.Summarizing the factors that may affect the data quality of participants,this paper proposes a vehicle score mechanism,and proposes three member optimization algorithms based on the vehicle score,which are the greedy selection algorithm,the degenerate greedy selection algorithm and the online member optimization algorithm.The results show that the quality of crowdsourcing in the degenerate greedy selection algorithm is the highest,and the time complexity of the greedy selection algorithm is small.The online member selection algorithm is suitable for high-definition map crowdsourcing scene with high real-time performance.Finally,this paper uses Luxemburg sumo traffic(LuST)data set to carry out simulation experiments.Experiments show that the proposed task publishing mechanism can guarantee the time coverage under different traffic flow,and the crowdsourcing utility can be increased by 8.34% by using the optimal stopping rule.The experiment of crowdsourcing member selection mechanism shows that the quality of crowdsourcing obtained by the degenerative greedy selection algorithm is the best among the three,followed by the greedy selection algorithm,but its time consumption is 79.3% less than that of the degenerative greedy selection algorithm.The online member selection algorithm can make the data uploaded 61.7% earlier on average,which is more suitable for high-definition map crowdsourcing scenarios with high real-time performance.Finally,the experiment also proves the effectiveness and necessity of the proposed region division algorithm based on perceptual features and the factors affecting the data quality of each participant. |
