Pedestrian detection data expansion method based on generative adversarial network

Abstract

The invention relates to a pedestrian detection data expansion method based on a generative adversarial network, and the method comprises the steps: S1, building a three-layer cascaded generative adversarial neural network model, and setting a target function of model training, wherein each layer of generative adversarial neural network adopts a Bicycle GAN structure, the generator adopts a residual Unet structure, and the input of the next layer of network is the pedestrian instance mask picture and the output of the previous layer of network; s2, preprocessing the training data; s3, traininga three-layer cascade generative adversarial neural network model by adopting the preprocessed data; and S4, completing expansion of pedestrian detection data through the three-layer cascade generative adversarial neural network model. Pedestrians generated by adopting the scheme of the invention are fused with the background more naturally, and details of the generated pedestrians are finer by improving the Unet structure of the generator; the multi-scale pedestrian picture is generated based on the cascade structure, so that the quality of the large-size and high-resolution pedestrian picture is improved; diversified pedestrians can be generated, and the data expansion efficiency is improved.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a pedestrian detection data expansion method based on a generative confrontation network. Background technique [0002] The invention relates to pedestrian detection, which is a basic task in video processing, and is widely used in scenarios such as intelligent video surveillance, automatic driving, robot automation, etc. Training a high-precision pedestrian detection model requires a large-scale, high-quality pedestrian picture data set. At present, research related to pedestrian detection mainly uses existing public data sets, most of which come from giant Internet companies, and they have invested a lot of manual labeling and correction costs to ensure the reliability of the data sets. When training models on these public datasets, researchers often use traditional data augmentation methods, such as image flipping, random cropping, and color adjustment, to augment the images in...

AI Technical Summary

Problems solved by technology

[0005] The technical problems to be solved by the present invention are: 1. Solve the problem of obvious edge traces when the pedestrian frame in the generated pedestrian picture is fused with the background; 2. Solve the p

Images