Street view images provide us with a lot of information about urban scenes, but preserving people's privacy in these images creates challenges. Looking for a solution to deal with these challenges and preserve people's privacy, we propose an automatic method to remove all pedestrians and vehicles from street view images. Our approach is able to detect and exclude all pedestrians and vehicles with high accuracy. In this approach, we use semantic segmentation with the help of 2DPriors to recognize people and objects. Using this method, semantic information with a rich structure is extracted from multimodal data including camera and lidar. This work improves performance in semantic segmentation. After detecting pedestrians and vehicles, we remove them using an inpainting network architecture that uses fast Fourier convolutions (FFCs) and a simple, one-step approach for large masks. Our privacy-focused approach provides the best performance in preserving people's privacy in street view images by applying advanced techniques in the detection and removal stages of pedestrians and vehicles.
