Research On Perceptual Information Processing And Fusion In Networked Autonomous Driving

With the rapid development of the automotive industry,people’s demand for cars is no longer limited to the functions of traditional cars.Autonomous driving and smart cars have become the focus of the industry.The safe driving of smart cars depends on the large scale of vehicle-side sensor network.How to ensure the effective processing and fusion of the sensory information of each node in the sensor network has a crucialimpact on the networked autonomous driving.Therefore,the fusion of node perceptual information and clustering among nodes are the main contents of this paper.In terms of multi-sensor information fusion in single node,how to fuse heterogeneous perceptual information perceived by different sensors on the node effectively is the first research direction of this paper.This paper first abstracts heterogeneous perceptual information into a series of vectors and maps them to autonomous driving decisions.Then,we propose an improved D-S(Dempster-Shafer)evidence theory to fuse abstracted perceptual data to reduce the computational resource and improve the accuracy of the fusion result.In terms of clustering of dynamic sensor network nodes,the autonomous driving area of the network is wide,the number of vehicle nodes is large,the distribution of nodes in the region is uneven and the network topology changes dynamically.This paper divides the entire networked autonomous driving area into multiple sub-areas(clusters),and then processes the perceptual information of each node in the sub-area.An interval-type fuzzy C-means clustering algorithm based on a weighted multi-width Gaussian kernel function is proposed to implement clustering of dynamic sensor network nodes.It speeds up the processing of the edge decision center for the perceptual information from the nodes,clustering more accurately to get more accurate processing results,and then feedbacks more reliable driving decision information to each vehicle node.