Real Time Target Detection for Infrared Images

The paper focuses on the real time detection and identification of human in IR images. Regular RGB cameras cannot capture images well in different weather conditions as dense fog, rain and complete darkness. IR objects due to its heat signature properties are well captured by IR cameras. YOLOv3 model is used as base model for IR detection as it delivers high frames per second.YOLOv3 is fine tuned for enhancing the performance and generalization ability of the model with IR images. The fine-tuned YOLO model is trained with customized dataset of IR images. The model achieves good accuracy of 89.28 mAP with validation set of IR images. The performance of tuned model is compared with the original YOLOv3 model. The original YOLOv3 model (without fine tuning) provides poor accuracy with 1.3 mAP with same validation set of IR images.