Accurate environment modeling method based on radar assistance

Abstract

The invention discloses an accurate environment modeling method based on radar assistance. The method comprises the following steps: 1, acquiring signals by a millimeter-wave radar; 2, obtaining depthand speed information of an object in a detection range; 3, enabling an optical sensor to extract a directed image; 4, performing semantic segmentation; 5, matching a target label with a millimeter wave point cloud target; 6, acquiring target category, distance and speed information. The invention provides a method for constructing a scene situation by fusing millimeter wave radar information andvisible light image information, which can not only align the same target in two sensor signals, but also construct an information vector belonging to a certain target; under the condition of large environmental interference (such as foggy days and rainy days), an accurate environmental model can be constructed through the cooperation of multiple sensors, so the driving safety is guaranteed to agreat extent, the intelligent level in the driving process is improved, and then the driving safety coefficient is improved.

Description

technical field [0001] The invention relates to an environment modeling method, in particular to an accurate environment modeling method based on radar assistance. Background technique [0002] With the development of intelligent driving technology, the understanding of the scene determines the height of its intelligence, so scene situation awareness has gradually become the top priority of intelligent driving technology. The construction of the scene model is a key technology in situation awareness. By marking the category, depth, speed and other information of the surrounding objects, it can help the intelligent module to carry out reasonable path planning, timely threat warning and avoidance operations. [0003] However, the current method of building a scene model has the following two problems: [0004] 1) The surrounding environment information model constructed by the current system is not accurate enough: there is an inaccurate target model problem, that is, for a c...

AI Technical Summary

Problems solved by technology

[0004] 1) The surrounding environment information model constructed by the current system is not accurate enough: there is an inaccurate target model problem, that is, for a certain target, the target category and position information provided in the optical image cannot correspond to the depth information of the same target in the millimeter wave radar conduct

The specific performance is that the position and category information of the target correspond to the depth and speed information inaccurately, that is, the information between the sensors has not been accurately fused and not aligned to the same target, resulting in the inability of the constructed model to efficiently cooperate with the artificial intelligence algorithm. Environmental situation awareness and intelligent auxiliary decision-making;

[0005] 2) At present, there is a lack of effective methods for information fusion of optical radar and millimeter-wave radar: the existing methods are all for in-depth information mining and analysis of single sensor data, which cannot build a high-precision scene model, and cannot integrate the advantages of multiple sensors for accurate fusion. The accuracy is poor, and the collected information is relatively simple

Moreover, most of the existing methods are based on traditional algorithms, which directly model by extracting statistical information of depth and gradient, which have relatively high requirements on the environment and sensors, poor robustness, and cannot work efficiently.

Images