Scene evaluation method and device, electronic equipment and storage medium

Abstract

The invention relates to a scene evaluation method and device, electronic equipment and a storage medium. The method comprises the steps that a test scene and a real scene are obtained, the scenes are continuous data sequences of the same target in the time dimension, the test scene is a scene obtained in a road test or a test field, and the real scene is a scene obtained after the test scene is marked; filtering the test scene and the real scene; matching the filtered test scene with a real scene; classifying the matched test scene and real scene; and performing calculation and statistics on the classified scene results to generate a result report. According to one or more embodiments of the invention, a better effect can be achieved with minimum time and labor cost.

Description

technical field [0001] The present invention relates to automatic driving, in particular to scene evaluation involved in automatic driving. Background technique [0002] At present, the L2 / L3 level of autonomous driving is mainly commercialized with assisted driving functions, such as adaptive cruise (ACC) / automatic emergency braking (AEB) / open door collision warning (DOW) and other functions. For the evaluation of these functions: 1) For the test cases in the test field, use manual frame-by-frame measurement and evaluation; 2) For the actual large amount of road test data, a limited number of common scenes and difficult scenes are selected for manual labor. Frame-by-frame measurement and evaluation, so as to obtain frame-by-frame accuracy rate and recall rate and other indicators, and feedback to guide the upgrade and optimization of the function. [0003] At present, the industry uses common manual methods to close the loop of testing and statistics, but this method consu...

AI Technical Summary

Problems solved by technology

[0003] At present, the industry uses a common manual method for closed-loop testing and statistics, but this method consumes a lot of time and energy of testers, and the coverage is not comprehensive

If the number of scenarios is large, a large amount of test mileage is required, and testers need to spend a lot of time and energy on repeated statistics and verification work, which is generally inefficient

At the same time, the common frame-by-frame statistical method obtains incomplete results and biased indicators, which cannot fully indicate the performance of the function, nor can it meet the evaluation requirements.

Images