A Deep Learning Based Approach for Detection of Face Mask Wearing using YOLO V3-tiny Over CNN with Improved Accuracy

Aim: The objective is to build an efficient face mask detector using YOLO V3-tiny. Materials and Methods: The algorithm used to detect face masks is novel YOLO V3-tiny in comparison with Convolutional Neural Network (CNN), the dataset used was (“Facemask Detection Dataset”) the sample size was 136. Results: Novel YOLO V3-tiny gets accuracy of 95% and for CNN it was 84%. On the basis of the network’s original two-scale prediction target, a scale is added to create a three-scale prediction, which can improve the accuracy of detecting small targets such as masks. The YOLO V3-tiny and CNN have a statistically significant independent sample t-test value (p0.001) with a 95 percent confidence level. Conclusion: face mask detection in YOLO V3-tiny has a significantly better accuracy than CNN.