Özet:

Computer vision algorithms are becoming more used with the advancement of technology. Although computer vision applications made with image processing and machine learning algorithms are classic methods, noise or unwanted environmental changes show the impact on the results. In this study, the detection of 4 geometric shapes in the bul-tak game was performed by two different methods. In the classic method, the properties obtained by image processing in two different algorithms were classified by the k-NN algorithm, and in the deep learning method, the Yolov4 algorithm was customized for object detection. In the experimental environment, the classic image processing method provides 100% success in the black background test data set, while the success in the second test data set with different colors and design backgrounds has fallen to 86.25%. The YOLOV4 deep learning method algorithm has achieved 100% success in both data sets. When algorithms are run on a real-time camera image, the classic method has performed the object detection process at 0.06 s in a black background square image, while in a square image with a different color and background design at 0.04 s, the Yolov4 method has performed the object detection process at 1.06 s.

Anahtar Kelimeler: