Automatic driving test scene analysis method and system

Abstract

The invention discloses an automatic driving test scene analysis method, and relates to the technical field of unmanned driving. Performing frame-by-frame information extraction and scene feature data analysis on a scene to be analyzed, performing similarity judgment on the scene feature data and the scene feature data stored in the database, and selecting the scene category with the highest similarity as the frame; screening out continuous multi-frame data belonging to the same scene category as a target scene fragment; calculating the difficulty of each frame of data in the target scene segment; and screening out continuous multi-frame data meeting the data difficulty requirement as a requirement scene fragment. According to the method, the target fragment is determined by adopting a scheme of scene similarity preliminary screening and difficulty quantitative evaluation, so that the problem that scene extraction and cutting cannot meet test requirements due to the fact that the scene cannot be divided carefully due to insufficient data volume in the early stage of development is avoided, and meanwhile, redundant scene fragments are effectively reduced.

Description

technical field [0001] The present invention relates to the technical field of unmanned driving, in particular to an automatic driving test scene analysis method and system. Background technique [0002] At present, most of the following two schemes are used for the processing of test scene data in autonomous driving scenarios. There are mainly two methods for data processing. One is to manually screen and cut, which is time-consuming and labor-intensive overall, and the efficiency is low; the other is Process the relevant feature quantities on the original data, and then analyze it based on the extracted features and the original scene category features, and use the start and end time points corresponding to the number of frames that meet the conditions (and the frames before and after) as the scenes that need to be retained The starting and ending points of the data, and based on this automatic slicing, the defect of this scheme is that if you need to divide and cut the au...

AI Technical Summary

Problems solved by technology

[0002] At present, most of the following two schemes are used for the processing of test scene data in autonomous driving scenarios. There are mainly two methods for data processing. One is to manually screen and cut, which is time-consuming and labor-intensive overall, and the efficiency is low; the other is Process the relevant feature quantities on the original data, and then analyze it based on the extracted features and the original scene category features, and use the start and end time points corresponding to the number of frames that meet the conditions (and the frames before and after) as the scenes that need to be retained The starting and ending points of the data, and based on this automatic slicing, the defect of this scheme is that if you need to divide and cut the automatic driving scenes in detail, you need to rely on the classification results of a large number of historical scene data sets, and at the same time, it is for the same category of scenes It is also impossible to quantitatively analyze the difficulty of the scene

Images