A Combined Method for Object Detection under Rain Conditions Using Deep Learning

The process of object detection utilizing deep learning is one of the most important deep learning applications and computer vision techniques, where one can learn image features in normal weather conditions and different rain conditions. Therefore, a deep convolutional neural network (DCNN) has become more important for object detection. Rain is a common and maj or factor in degrading image quality and decreasing object detection reliability. The main aim of this work is to remove rain streaks to get high reliability in detection process and decrease the error rate, in normal conditions and different rain conditions (light, medium and heavy). Firstly, the quality of the images is improved and removed rain streaks by de-raining algorithm that use the Deep Detail Network (DDN) method. Then the way deep learning is the main object detector, through use the YOLO to detect objects and determine its type. YOLOv3 and tiny-YOLOv3 have been determine from the literature review as the most suitable and efficient algorithms for detecting objects in real time after improving the image quality. The performance of these algorithms has been calculated and compared with each other. To evaluate the effectiveness of the devised approach (De-raining+YOLOv3), Fl-score, Recall, and Precision were computed. Using the proposed method combined from DDN with YOLOv3 technique (De-raining+YOLOv3), the mean of Fl-score of 95.02%, Recall of 97.22%, and a Precision of 92.92% were attained. Our presented approach is more resilient and accurate in object detection under rain conditions according to the findings of the results of the experiments. It is considered the best way for object detection under rain conditions with high reliability.