基于U-Net神经网络的行人图像语义分割

Pedestrian Image Semantic Segmentation Based on U-net

针对图像中人体部分的精确分割问题,本文提出一种基于U-Net神经网络的行人图像语义分割方法。该方法首先进行四次卷积和最大池化处理,实现了下采样,提取出图像中的行人特征;其次进行四次卷积和反卷积处理,实现了上采样;最终通过卷积获得了图像分割结果。使用戴姆勒行人检测标准数据库训练神经网络并进行测试,分割效果良好。

Aiming at the problem of accurate segmentation of human part in image, a semantic segmentation method of pedestrian image based on u-net neural network is proposed in this paper. Firstly, the method performs quartic convolution and maximum pool processing to realize down sampling and extract pedestrian features from the image;Secondly, the up sampling is realized by quartic convolution and deconvolution;Finally, the image segmentation results are obtained by convolution. Using Daimler pedestrian detection standard database to train the neural network and test it, the segmentation effect is good.