Application of Transfer Learning in Infrared Pedestrian Detection

Abstract

Object detection is one of the most important and challenging branches of computer vision, which has been widely applied in our life. With the rapid development of deep learning networks for detection tasks, the performance of object detectors has been greatly improved. However, the lack of large labeled dataset obstructs the usage of convolutional neural networks (CNN) for detecting in thermal infrared images. Most existing dataset focus on visible images, while thermal infrared images are helpful for detection even in a dark environment. To address this problem, we propose the use of transfer learning to improve the accuracy of infrared pedestrian detection. We pretrain a convolutional neural network on a large dataset (which contains 1.8 million images with 654 categories), then use the convolutional neural network as a fixed feature extractor for the task of infrared pedestrian detection. The average precision of detection using ImageNet pretrained model alone is 83.34%. By adding ours pretrained model, the average precision has improved to 84.78%. We believe that the method of transfer learning can be extended to other infrared detection applications and achieve other breakthroughs.