Target scene simulation model construction method, data set generation method and electronic equipment

Abstract

The invention discloses a method for constructing a simulation model of a target scene, a method for generating a data set, and an electronic device. The method includes performing hierarchical segmentation of the target scene according to texture scales to obtain multiple levels of segmentation objects; for the multiple segmentations object, constructing a simulation model corresponding to each object; acquiring image data of a target scene; segmenting a low-resolution texture image corresponding to each segmented object from the image data of the target scene according to the type of the segmented object ;Synthesize the low-resolution texture image corresponding to each segmentation object to obtain the high-resolution texture image corresponding to each segmentation object; respectively map the high-resolution texture image corresponding to each segmentation object to the corresponding In the simulation model of the simulation model, a simulation model with texture is obtained; the simulation model with texture corresponding to each object is unified into the scene, and the simulation model of the target scene is obtained. It is suitable for the generation of large-scale binocular datasets in scenarios such as extraterrestrial detection and automatic driving.

Description

technical field [0001] The present application relates to the technical field of computer vision, and in particular, to a method for constructing a simulation model of a target scene, a method for generating a data set, and an electronic device. Background technique [0002] The method of obtaining 3D depth information by stereo matching with binocular cameras is a hot research topic in the field of computer vision technology. Compared with the traditional binocular stereo matching algorithm, the neural network-based binocular stereo matching algorithm is showing increasingly excellent results in terms of algorithm accuracy and operating efficiency, and the neural network algorithm needs to rely on high-quality large-scale binocular stereo matching. Deep dataset. [0003] At present, the mainstream real scene datasets in the field of binocular stereo matching research mainly include the KITTI dataset that uses lidar to collect depth information, and the Middlebury dataset t...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the embodiments of the present application is to provide a method for constructing a target scene simulation model, a method for generating a data set, and an electronic device, so as to solve the problems in related technologies that the texture of the simulation model of the data set is quite different from that of the real scene, and the scale of the real scene data set is not large. Due to the technical problems of large size, limited precision and density, this method makes full use of the scale and distribution characteristics of texture information to obtain a large-scale binocular depth data set that is closer to the real scene, and meets the training and application requirements of the binocular stereo matching neural network.

Images