Pedestrian detection method and apparatus for monitoring based on image analysis

Abstract

The invention relates to a pedestrian detection method and device for monitoring based on image analysis, in particular to a method for learning a pedestrian detector to be used for robust surveillance or military purposes based on image analysis is provided for a solution to a lack of labeled images and for a reduction of annotation costs. The method can be also performed by using generative adversarial networks (GANs). The method includes steps of: generating an image patch by cropping each of regions on a training image, and instructing an adversarial style transformer to generate a transformed image patch by converting each of pedestrians into transformed pedestrians capable of impeding a detection; and generating a transformed training image by replacing each of the regions with the transformed image patch, instructing the pedestrian detector to detecting the transformed pedestrians, and learning parameters of the pedestrian detector to minimize losses. This learning, as a self-evolving system, is robust to adversarial patterns by generating training data including hard examples.

Description

technical field [0001] The present invention relates to a learning method and a learning device, a testing method and a testing device used with an autonomous vehicle, more particularly, a pedestrian detector used in Robust Surveillance based on image analysis using GAN ( A learning method and a learning device of Pedestrian Detector, a testing method and a testing device using the learning method and learning device. Background technique [0002] Convolutional Neural Network (CNN or ConvNet) in Machine Learning is a category (Class) of Deep Feed-Forward Artificial Neural Network (Deep, Feed-Forward Artificial Neural Network) successfully applied to visual image analysis . [0003] Such a CNN-based object detector (i) makes at least one convolutional layer apply a convolution operation to the input image to generate a feature map corresponding to the input image, (ii) makes the RPN (Region Proposal Network: Region Generation Network) utilize the feature Figure generates ca...

AI Technical Summary

Problems solved by technology

[0005] However, conventional pedestrian detectors have the following problems: Pedestrians with unique shapes and / or patterns that have not been encountered in the training data, pedestrians wearing black clothes on dark roads, and pedestrians similar to the surrounding background Pedestrians cannot be detected accurately in cases such as

[0007] However, there are disadvantages as follows: it is impossible to prevent detection failures by periodically re-educating the pedestrian detector, and whenever a detection failure occurs, supplementary re-education is required, and monitoring is required. Use of extra manpower undetected

[0008] Also, it is difficult to secure appropriate training data for re-education on detection failure cases

Images