Human Detection in crowd using One-stage and Two-stage object detection models

Object detection has various uses for recognition, detection, and tracking. Detecting humans in crowded low-quality images is an emerging topic and is a difficult task. In this paper, different object detection are compared for our purpose. Here we use the two models - YOLO and Faster R-CNN and test their working for human detection on low-quality images. We take a low-quality image collection and utilize pre-processing techniques to improve its quality. Using the same experimental setting, we build YOLO and Faster R-CNN and evaluate their results based on accuracy and speed. After comparing the results it shows that in terms of detection speed, YOLO outperforms Faster R-CNN. And YOLO lags than Faster R-CNN when it comes to accuracy. Further we also discuss the advantages and disadvantages of using these models and which is best suited for the scenario we intend to work on. Our findings give useful insights for the development of low-quality image object detection systems for accurate human detection in crowded images, with practical implications for a variety of applications.