Object Detection Based on Fusion of Visible and Infrared Images

Abstract

In consideration of the complementary characteristics between visible light and infrared images, this paper proposes a novel method for object detection based on the fusion of these two types of images, thereby enhancing detection accuracy even under harsh environmental conditions. Specifically, we employ an improved AE network, which encodes and decodes the visible light and infrared images into dual-scale image decomposition. By reconstructing the original images with the decoder, we highlight the details of the fused image. Yolov5 network is then constructed based on this fused image, and its parameters are adjusted accordingly to achieve accurate detection of objects. Due to the complementary information features that are missing between the two image types, our method effectively enhances the precision of object detection.