Research On The Identification Method Of Electric And Fuel Cars Based On UAV Thermal Infrared Images

The development of electric vehicles helps to reduce carbon emissions and environmental pollution.The number and proportion of electric vehicles and fuel vehicles on the road will help the government to grasp the situation,assist the deployment of electric vehicle charging piles,and study the impact of vehicle emissions on the urban environment.Due to the existence of hybrid vehicles and some other reasons,the identification method of electric vehicles and fuel vehicles based on visible images has great defects.Taking small cars as an example,this paper proposes to use UAV thermal infrared remote sensing imaging to identify electric vehicles and gasoline vehicles on the road.Use the UAV equipped with a dual sensor thermal solution to image the tail of the car target on the road in a tilted manner,and generate a dataset image after preprocessing.Establish a visual interpretation rule based on the heat flow of the car’s tail when the dataset is labeled to distinguish electric cars and fuel cars.A semiautomatic labeling method using target tracking algorithm to update target position is used in labeling to improve labeling efficiency.This labeling method can be applied to the labeling process of other similar scenes.After the above steps,a thermal infrared grayscale dataset is generated.The thermal infrared-visible fusion dataset is produced by data fusion of the visible image and thermal infrared image obtained at the same time as the thermal infrared data.The temperature data is derived from the original thermal infrared data and reconstructed into an image to make the thermal infrared temperature data set.These three datasets share the same annotation files,the difference is that the images are different,and the three datasets produced will be used in the subsequent experimental process.Comparing our proposed semi-automatic annotation method with the STC target tracking source code method,the results show that our algorithm has the highest Io U accuracy.Using YOLOv5 and SSD algorithms to identify the thermal infrared grayscale dataset,the m AP result reaches 0.99.Experiments on the thermal infrared-visible fusion dataset show that the accuracy of the recognition of electric cars and fuel cars is reduced,but the recall rate of small targets is higher,which proves that the fusion of thermal infrared images with visible images does not always improve the accuracy,however,the detection accuracy of small targets can be improved by increasing the image resolution.The model trained with the thermal infrared temperature dataset and without color enhancement is somewhat sensitive to temperature.Identify electric and fuel cars on the road using a trained model,and use the target tracking algorithm to make statistics to obtain the proportion of electric cars in different time periods,which reflects the travel time rules of the electric cars group to a certain extent,indicating that this study also has potential application prospects.