Motion Target Detection Based On Fused Images

The technology of fusing infrared and visible light dual-mode images combines the thermal imaging mode of infrared images with the reflective imaging mode of visible light images.It effectively compensates for the limitations of single-mode imaging in certain scenarios,thus obtaining more authentic scene information in the images.Object detection,as a vital research direction in computer vision,plays a crucial role in various research domains.The objective of this study focuses on the detection of objects using the fusion of infrared and visible light dual-mode images,which exhibits superior detection performance compared to single-mode images.Therefore,this research topic holds significant value.In this paper,we enhance image quality through the fusion of infrared and visible light images and conduct research on dynamic object detection based on fused images and dynamic object detection systems using the fusion of infrared and visible light dual-mode images.Initially,this paper conducts a research analysis on the fusion methods of infrared and visible light dual-mode images and proposes an image fusion method.By employing an improved low-rank representation method,we decompose the infrared and visible light images into their basic layer components and salient layer components separately.Corresponding fusion methods are applied to the fusion stage based on the characteristics of different layer information.The fused image obtained exhibits superior performance in objective evaluation metrics such as average gradient and boundary intensity.Visually,the fused images generated by our proposed method demonstrate certain resistance to interferences like lighting and smoke,and provide more reliable image information in terms of scene reconstruction.Subsequently,building upon the research of dual-mode image fusion,we delve into the study of dynamic object detection algorithms.We improve two detection algorithms based on different computational capabilities.In cases where computational resources are limited,we propose an improved three-frame differencing method by combining frame difference technique with logical operations.The improved three-frame differencing method achieves higher accuracy in contour boundaries of dynamic objects.For situations with abundant computational resources,we introduce depth separable convolutions to replace the original convolutional network in the existing Center Net network and incorporate attention mechanisms into the network structure.We train the improved network on a created dataset.Furthermore,we conduct experiments and evaluations on the detection algorithms,validating the superior performance of dual-mode fusion image detection compared to detection on single-mode images.We also compare the detection performance of various algorithms on dual-mode fusion images.Finally,this paper integrates the fusion of infrared and visible light dual-mode images with object detection and conducts research on system design.A target detection system based on the fusion of infrared and visible light dual-mode images is designed.Through experimental testing,the system achieves object detection in low-light conditions,providing a method for object detection in complex scenes.