LEARNING METHOD AND LEARNING DEVICE OF PEDESTRIAN DETECTOR FOR ROBUST SURVEILLANCE BASED ON IMAGE ANALYSIS BY USING GAN AND TESTING METHOD AND TESTING DEVICE USING THE SAME

摘要

The present invention relates to a method for learning environmentally unaffected surveillance (Robust Surveillance) or a pedestrian detector for testing used for military purposes based on image analysis. (Annotation) Provided for cost reduction, can be performed using GAN (Generative Adversarial Network), crop each of the areas on the training image, create an image patch, and hostile style converter (Adversarial) Style Transformer) to generate a transformed image patch by transforming each pedestrian into a transformed pedestrian that can make detection difficu And generating a transformed training image by replacing each of the regions with a deformed image patch, allowing the pedestrian detector to detect the deformed pedestrian, and learning the parameters of the pedestrian detector to minimize loss. Characterized in that, this learning as an autonomous evolution system (Self-Evolving System) is characterized in that it may not be affected by the adversarial pattern by generating training data including difficult examples.