Network driving environment integrated perception model based on convolutional and hollow convolutional structure

Abstract

A network driving environment integrated perception model based on a convolutional and hollow convolutional structure simultaneously realizes object detection and semantic segmentation. A video imageof a road environment is shot through a forward-looking camera system mounted on a vehicle. The residual network model is used to get the bottom feature map of the image. The converged network is designed, which includes two sub-modules: object detection and semantic segmentation. The two modules share the bottom feature map. Among them, the object detection module is responsible for predicting the confidence level of the object frame and the category, and the semantic segmentation module is responsible for predicting the pixel level of each category. The appropriate loss function is selectedfor each of the two modules, and the perceptual model tends to converge in both modules after alternate training. Finally, the joint loss function is used to train the two modules simultaneously to get the final perceptual model. The model can simultaneously complete object detection and semantic segmentation with small computation amount, and the perceptual model uses a large amount of data of object detection to assist the semantic segmentation module to learn the image distribution law.

Description

technical field [0001] The invention relates to the technical field of advanced automobile driver assistance, in particular to a fusion network driving environment perception model based on convolution and atrous convolution structures. Background technique [0002] Driving environment perception is an important function of ADAS (Advanced Driver Assistance System). The existing driving environment perception mainly includes object detection (for objects of interest, such as pedestrians, vehicles, bicycles, traffic signs, etc., to obtain the position information and category information of the object in the image) and semantic segmentation (for each pixel of the image Dots mark categories respectively) two major tasks. Driving environment perception can be used to assist driving decision-making and reduce traffic accidents. [0003] At present, in order to complete target detection and semantic segmentation, statistical learning methods such as support vector machines or co...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the above problems, the present invention provides a fusion network driving environment perception model based on convolution and atrous convolution structure, which solves the problem that the current driving environment perception model has a large amount of calculation, repeated calculations, a single-task model solves a single problem, and the semantic segmentation model does not Semantic segmentation data set requirements are too high (the cost of pixel-level data labeling is too high), and the problem of multi-task driving environment perception cannot be completed at the same time. To achieve this goal, the present invention provides a fusion network driving environment based on convolution and hole convolution structure Perceptual model, concrete steps are as follows, it is characterized in that:

Images