Pengenalan Personal Menggunakan Citra Tampak Atas pada Lingkungan Cashierless Strore

Abstract

Personal recognition with image processing techniques from the side view has the disadvantage of being applied to the cashierless store environment, namely inaccurate recognition or identification when personal collisions occur. To overcome this, the image capture method is used from the top-view. Personal recognition method through the top-view image using the Haar Cascade Classifier method. 1420 positive images and 2170 negative images are used to find features that are considered suitable for recognizing objects using the Adaptive Boosting (Adaboost) method. Tests were carried out on 100 test data by varying the parameters of min_neighbors (3.4, and 5) and the size of the dataset window (25x25, 35x35, 45x45 pixels). Personal recognition testing gets the highest accuracy of 89.9% with the parameters used are min_neighbors 5 and the size of the 25x25 pixel dataset in the detection parameter size of min_size 140x140 pixels.