Method and device for constructing high-precision map

Abstract

The invention discloses a method and a device for constructing a high-precision map. The method comprises the steps of performing semantic sensing on one or a plurality of images of an on-vehicle camera from a vehicle based on a deep learning algorithm; performing three-dimensional reconstruction on an attention road sign according to vehicle attitude data and a semantic sensing result; parameterizing a three-dimensional reconstruction result; and adding a parameterizing result into a high-precision map database. The method and the device according to the invention have advantages of greatly reducing manufacturing cost of the high-precision map, facilitating data acquisition mode in a crowdsourcing manner and increasing update frequency of the high-precision map.

Description

technical field [0001] The present disclosure generally relates to the technical field of automatic driving, and in particular relates to a method and device for constructing high-precision maps in a low-cost manner. Background technique [0002] Autonomous driving often requires the use of high-resolution maps. Such high-precision maps can achieve centimeter-level accuracy and contain richer detailed information (e.g., road slope, curvature, etc.) than digital maps, which are crucial for vehicle navigation, localization, control, and safety. [0003] For high-precision maps, it is usually necessary to use expensive lidar and high-precision positioning sensors to collect point cloud data, and then professional technicians perform a series of operations on the point cloud data such as data cleaning, splicing, alignment, data vectorization, and indexing , so the cost of equipment and the requirements for the skill level of professionals are very high, which makes it difficult...

AI Technical Summary

Problems solved by technology

[0003] For high-precision maps, it is usually necessary to use expensive lidar and high-precision positioning sensors to collect point cloud data, and then professional technicians perform a series of operations on the point cloud data such as data cleaning, splicing, alignment, data vectorization, and indexing , so the cost of equipment and the requirements for the skill level of professionals are very high, making it difficult to deploy high-precision maps on a large scale, and greatly limiting the update frequency of high-precision maps

In addition, every link of the usual high-precision map production process requires manual participation, which also greatly limits the production efficiency of high-precision maps

[0004] In addition, in the production process of the usual high-precision map, in order to obtain the structure and color information of the scene at the same time, it is often necessary to use the camera and the lidar sensor at the same time. Increased system and process complexity

[0005] In addition, the high-precision map made in the usual way may contain a large amount of redundant information or potential dynamic objects that are currently stationary but actually moveable, thus increasing the pressure on the storage and transmission of the high-precision map, thus extremely The earth limits the update efficiency of high-precision maps

Images