Real-time streetscape image semantic segmentation method based on staged feature semantic alignment

Abstract

The invention discloses a real-time streetscape image semantic segmentation method based on staged feature semantic alignment, and relates to a computer vision technology. Firstly, a lightweight image classification network ResNet-18 and an efficient space-channel attention module are used for constructing an encoder, and a plurality of feature alignment modules with different designs and a global average pooling layer are used for constructing a decoder. Thirdly, forming a semantic segmentation network model based on an encoder-decoder network structure by using the obtained encoder and decoder; and finally, the features in the encoder and the output features of the decoder are aggregated and sent to a semantic segmentation result generation module to obtain a final semantic segmentation result. A corresponding segmentation result can be efficiently generated at a real-time rate without lowering the image resolution while maintaining a high-resolution input image. Compared with an existing real-time semantic segmentation method, the method has the advantages that more excellent segmentation precision can be achieved, and better balance is achieved between speed and precision.

Description

technical field [0001] The invention relates to computer vision technology, in particular to a semantic segmentation method for real-time street view images based on semantic alignment of staged features. Background technique [0002] Semantic segmentation is one of the key technologies for scene understanding. It needs to predict each pixel in the image to realize the pixel-level semantic category classification of the image. In recent years, the application of autonomous driving and intelligent transportation has attracted widespread attention. In these applications, an urgent problem to be solved is how to provide a comprehensive understanding of traffic conditions at the semantic level. Therefore, it is extremely important for these applications to study the semantic segmentation method of street view images and provide pixel-level street view scene understanding. [0003] In recent years, benefiting from the development of convolutional neural networks, a large number...

AI Technical Summary

Problems solved by technology

Although these methods greatly reduce the computational complexity of semantic segmentation, they lose contextual information or spatial details to a certain extent, resulting in a significant drop in accuracy.

Images