A KNN Query Method for Autonomous Driving Sensor Data

Abstract

Autonomous driving cars need to perceive and extract environmental information through sensors around the body during the driving process. Sensor data streams are rich in semantic information, and semantic annotation can form semantic text with geographic location information. This information is important for scenario applications such as effective updating of high-precision maps and more accurate perception of the surrounding environment. Therefore, it is necessary to store and organize them effectively. At the same time, users’ demand for such information retrieval is increasing, and it is more and more important to efficiently retrieve useful information for users from the huge amount of information.In this paper, We propose an index, SIR-tree, to efficiently organize the spatial, textual and social information of objects. SIR-tree is an R-tree containing both social relevance and textual relevance, and each parent node in the tree contains the spatial, textual and social information of all children nodes, and has good scalability. Based on the SIR -tree index we propose the priority traversal algorithm BF, which uses a priority queue to store all candidate objects and traverses the nodes to filter them according to their ranking function values. To optimize the query algorithm, we propose two pruning strategies, distance-based BF algorithm and X-HOP_BF algorithm, to improve the query efficiency. Experimental results show that the BF algorithm takes at least 70 s to return 20 objects in the Gowalla dataset, while the X-HOP_BF algorithm only takes about 20 s.