Object Detection in Thermal Images using Dw-Yolov8 Under Low Light and Foggy Conditions

Object detection in thermal images is essentially used in the applications such as night time surveillance, search and rescue operations, military systems where the RGB cameras are fail to recognise the objects under low light and challenging weather conditions. The YOLO model is widely used for the object detection under these situations. In this paper, we propose a object detection framework for thermal images under low light conditions by modifying the YOLOv8 model by replacing the convolution layer with the depth wise convolution (DW Conv).By incorporating the DW conv layer in yolov8 will reduce the number of computations and also increase the accuracy of the yolov8 model. By replacing the convolution with DW conv the number of parameters and Floating-Point operations (FLOPS) will lead faster interpretation for detection of objects. The proposed model is experimented on the FLIR ADAS thermal dataset. The proposed model performance is analyzed qualitatively and quantitatively by obtaining the precision, recall, mAP, FLOPs, model parameters. And the results are achieved better performance and reduced computations when compared with the standard YOLOv8 model.