Abstract: - In the face of the global Covid-19 scenario, the process of softening the curve of the corona virus will be difficult if citizens do not take steps to prevent the spread of the virus. With no vaccine available, social distancing is the only possible way to combat the epidemic. The proposed framework uses the YOLO v3 object detection model to identify people in the background and in-depth tracking of identified people with the help of binding boxes and assigned IDs. The model results of YOLO v3 are compared to other popular modern models, e.g. CNN-based regional speed (convolution neural network) and single-shot detector (SSD) in terms of average accuracy (mAP), frames per second (FPS) and loss values are defined by object classification and location. Later, the L2 line shown in pairwise is calculated based on the three-dimensional feature space obtained using links and the size of the binding box. The name of the infringement index is proposed to reduce the inconsistency of the public deviation process. From the experimental analysis, it is evident that YOLO v3 with an in-depth tracking scheme shows good results with moderate mAP and FPS score to monitor community deviations in real time. We are using the YOLO v3 object acquisition model and the OpenCV image processing library to run this project. The project will play an important role in an area where large numbers of people can be expected such as a shopping mall or movie theater or airport. With the help of this project we can ensure that people follow the process of socialization.
Keywords: YOLO v3, Covid-19, Social Distancing, Pretrained Model, Webcam, CNN.
|
DOI:
10.17148/IJARCCE.2021.10627
