Research On Thermal Infrared Image Reflection Detection Algorithm Based On Target Detection

Thermal infrared imaging technology is widely used in industrial production,military drills,and everyday life because of its non-contact,relatively accurate capture of invisible temperature differences.However,infrared light has the characteristics of strong reflectivity,poor penetrating power,and strong environmental influence,which also makes thermal infrared reflection have the strongest impact on image quality.The purpose of this paper is to effectively detect and locate reflections in thermal infrared images.Based on the analysis of related detection algorithms under natural light conditions,the thermal infrared image reflection detection algorithm based on deep learning is mainly studied,and an effective thermal infrared pedestrian reflection detection algorithm is proposed.Based on the proposed thermal infrared image reflection detection algorithm,combined with natural light,the algorithm is extended from single pedestrian reflection detection to general object thermal infrared reflection detection.The main work and innovations of this paper are as follows:(1)A thermal infrared pedestrian reflection detection algorithm based on parallel target detection is proposed.Aiming at the problem of less available information in infrared images,this paper combines deep learning methods and makes full use of the ability of deep learning to mine information,and designs a pedestrian infrared reflection detection method based on parallel target detection.Because the reflection is generated by pedestrians,the algorithm adopts the idea of finding the pedestrian first and then the reflection.That is,two target detections are performed on a thermal infrared image at the same time,the first time targeting pedestrians,and the second targeting pedestrians and reflections.Then,the location relationship between the two output detection boxes is used to map the pedestrian detection box to the combination detection box.Finally,the corresponding detection box is subtracted to obtain the final pedestrian reflection detection box.Experimental results show that the algorithm can effectively detect the position of reflections from thermal infrared images containing pedestrian reflections and eliminate unnecessary interference.(2)An infrared universal reflection detection algorithm combining natural light and thermal infrared is proposed.There are two main difficulties in extending the reflection detection of thermal infrared images from pedestrians to general objects.First,most objects that can produce reflections tend to have relatively consistent temperatures and always show uniform brightness and similar shapes in thermal infrared images,so it is difficult to distinguish them by infrared images.Second,objects other than pedestrians are often not placed on the reflecting surface,and there is a distance between the real image and the reflection.In this paper,based on the pedestrian reflection detection method of thermal infrared image based on parallel target detection,a general infrared reflection detection algorithm combining natural light and thermal infrared is proposed.Aiming at the first difficulty of general reflection detection of thermal infrared,natural light image is added to the detection,so that the object can be effectively detected.In response to the second difficulty,a refinement step was added after the reflection positioning to make the detection results more accurate,effectively retaining the advantages of the thermal infrared pedestrian reflection detection method and promoting it as a general algorithm.Experimental results show that the algorithm can effectively detect thermal infrared reflections of various objects.