Pedestrian Detection Based on Infrared Imaging Through Gray Transformation and Deep Learning

Infrared images are widely used in night security. Compared with visible images, infrared images have lower pixels and poor imaging quality. The accuracy of pedestrian detection will be extremely poor. Considering these complex situations when the ambient temperature is high or complex, we combine the object detection with the unique temperature information of infrared image and propose a better method that can significantly improve the detection accuracy of infrared image in above situations. We carry out a variety of different preprocessing of the image and then carry out pedestrian detection of the processed image. This method has achieved a higher recognition accuracy in our own infrared image dataset. This will be applied in night security automation.