Research On Face Temperature Measurement Technology Based On Bispectral Fusion

In recent years,domestic and foreign countries have been attacked by the new coronavirus,which seriously affects social stability,economic development and people’s life safety.One of the main characteristics of infection with this virus is abnormal body temperature,so in public places with high traffic flow,fast and accurate screening of people with abnormal body temperature is the top priority for epidemic prevention and control.At present,the more commonly used temperature measurement methods in epidemic prevention work require epidemic prevention personnel to operate the frontal thermometer and infrared thermal imaging camera for temperature detection at close range,which has a large measurement error and is prone to misdiagnosis and cross-infection risk.Although the existing face temperature measurement system has the advantages of rapid temperature measurement and wide measurement range,it can only accurately measure temperature in the restricted area due to the limitation of the external black body.To address the shortcomings of the existing face temperature measurement system,this paper selects the faces of moving people as the target and investigates the key technology of the non-contact infrared human temperature measurement system,using dual-spectrum fusion face temperature measurement technology to achieve accurate and rapid detection of the surface temperature of the target face without external black bodies,thus providing a technical guarantee to overcome the new crown epidemic.1)The lightweight improvement of face detection algorithm is realized.To address the problem that face detection algorithms are not easily deployed on embedded platforms,an improved lightweight face detection network is designed to guarantee the detection accuracy while also improving the detection speed.By using k-means++ to re-cluster the face dataset to obtain anchor frames that are more suitable for the size of faces in pictures;for the problem that small target face samples in the dataset are easily missed,the Focal loss function is used to replace the classification loss part of the original loss function.2)A fusion registration scheme of binocular distance measurement and image based on face detection results is proposed.For the first step,based on the face detection result,the face detection frame is used as a feature point to match and calculate the parallax of the target image based on the distance from the face to the lens to be measured in real time.For the second step,the face distance information can be calculated by combining the binocular ranging principle,and the relative error of the ranging accuracy is within 1.5%,which meets the industry requirements;For the third step,based on the idea of point feature alignment,the face detection frame and the spatial transformation model are used to perform the pixel-level binocular ranging and image fusion alignment.Finally,based on the idea of point feature alignment,the pixel-level bispectral image fusion alignment is performed to determine the face region in the infrared thermal image.The experimental results show that the proposed method is highly accurate and easy to be deployed in embedded devices.3)Establish dynamic temperature compensation model.In the first place,the main factors affecting thermal infrared temperature measurement are analyzed,and temperature compensation is performed in terms of both temperature measurement distance and ambient temperature;In the second place,a dynamic distance temperature compensation model is established by combining binocular ranging technology,and ambient temperature is collected for ambient temperature compensation,which improves the overall temperature measurement accuracy.Finally,this paper builds an experimental platform for dual-spectrum fusion face temperature measurement technology.Based on the key technologies of the face temperature measurement system fusing visible and thermal infrared images,the experimental platform of the dual-spectrum fusion face temperature measurement technology is built and the experiments are conducted in an indoor environment.The experimental results show that there is a high detection rate and accuracy when detecting the body temperature of people in public places with large flow of people,and the error of the temperature measurement method is within 0.3℃ without external black body,which can meet the demand of rapid screening of abnormal body temperature.