Smart Detection Under Different Weather Conditions

Abstract

Object detection is one of the most essential and challenging tasks in computer vision and deep learning. The main goal of object detection is to determine whether the image has an object from predefined categories and then to return the class and spatial location of that object. Researchers achieved a significant improvement in object detection in both speed and accuracy due to the ability to learn from raw pixels. There are three main stages in object detection: region proposal, feature extraction, and classification. The current state-of-art object detection algorithms are divided into two categories: two-stage and one-stage. The two-stage algorithms perform the first two stages separately, while the one-stage algorithms perform these two stages together. A two-stage algorithm like faster R-CNN is known for its superb accuracy, while the one-stage algorithms like YOLO and SSD are much faster than one-stage algorithms. Still, they lack accuracy, especially with a small object. This work targeted the accuracy, so the two-stage detection algorithms, faster R-CNN, were adopted as the basic structure for the detection network, historically evaluated under different weather conditions. The study implemented and tested the faster R-CNN with VGG16 as a feature extractor with images under differing weather conditions. First, the study trained the network under different training parameters to obtain the best detector. Then, the study tested and evaluated the two best detectors under different weather conditions. The results show that the accuracy of the dete