Automatic Hand Recognition Using Deep Learning and YOLOv8

The identification of human hands is a critical step in the pre-processing stage of most computer vision tasks. This is because human hands are involved in the action. Estimating the hand posture, recognizing hand gestures, monitoring human activities, and other similar jobs are some examples of the kinds of work that fall under this category. Human hands are capable of a broad range of mobility and can alter their outward appearance in a great number of distinct ways. Because of this, it might be difficult to distinguish certain hands in a crowded area, and the movements of some hands can take a wide variety of forms. In this inquiry, we give a condensed examination of CNN-based object recognition algorithms; more specifically, we focus on the Yolov8n models with both 50 and 100 epochs. This study investigates a wide variety of object detectors, some of which are employed in the process of locating hand identification applications. We also use the Yolov8 models with the Oxford Hand Dataset for training and testing our suggested method. The amount of GFLOPS, the mean average precision (mAP), and the detection time are only a few examples of the metrics used to track and monitor performance. When compared to other approaches, the findings of our research suggest that Yolov8n's training phase, which consists of 50 epochs, is the strategy that offers the greatest degree of stability. During the training process, it was able to attain a precision rate of 77.6%, a recall rate of 68.7%, and a mAP of 75.8%. During the testing phase, Yolov8n achieved the maximum possible average score on the mAP, which was 75.8%.