Generative adversarial network-based algorithm for 3D construction of pedestrians

In response to the problem of poor recognition effects of existing pedestrian 3D construction algorithms, this paper proposes a GAN-based pedestrian 3D construction algorithm. By conducting in-depth research on the GAN model and optimizing the generator and discriminator networks in the model, a pedestrian learning model is constructed. We have introduced appearance depth features and motion depth features to improve the accuracy of the generator network's output images. In addition, we have introduced real human parameter set data into the discriminator network. By using correlation discrimination with the output images of the generator network, we can obtain 3D images that are close to real. The experimental results show that the algorithm has high accuracy and stability, and can be applied in multi-person environments.