The Research On Real-time Semantic Segmentation Method For Autonomous Driving Environment Perception

As one of the important technologies in the field of artificial intelligence,semantic segmentation aims to predict and assign class labels to all pixels in the image.It has broad application prospects in the field of automatic driving.This paper studies the real-time semantic segmentation method for automatic driving environment perception.The main research contents are as follows:(1)A lightweight feature extraction backbone network is designed,and a lightweight multi-scale feature aggregation module is constructed.Combined with spatial attention and channel attention mechanism,a double attentionguided concatenate module is obtained.A lightweight dual attention real-time semantic segmentation network is established,and the parameters of the network is 7.11 M.The test results based on Cityscapes dataset show that the MIo U is 64.9% and the segmentation speed is 50.2FPS,which verifies the effectiveness of the network in balancing the segmentation accuracy and speed.(2)In order to better preserve the spatial information,the extraction process of spatial features and semantic features are separated.A spatial feature extraction link including three convolutional downsampling layers is constructed,and a semantic feature extraction link is constructed through a lightweight multi-scale feature aggregation module,a lightweight attention module and a backbone network.Combined with the cross aggregation module to establish a lightweight two-way real-time semantic segmentation network,and the parameters of the network is 6.52 M.The test results based on Cityscapes dataset: MIo U is 67.8%,segmentation speed is 79.9FPS;The test resultsbased on Camvid dataset: MIo U is 64.2%,segmentation speed is 60.7FPS.