Turkish Online Journal of Qualitative Inquiry (TOJQI) Volume 12, Issue 6, June 2021: 1662-1666

Abstract

Recently, the performance of pedestrian recognition techniques has been improved rapidly with the introduction of deep learning, and the scope of their use has also been widely used in various fields. Currently, most pedestrian recognition technology focused on is detecting a standing person. These general pedestrian recognition techniques are inappropriate for situations where detailed pedestrian information is needed, such as complex traffic conditions, child protection zones, and disasters that make it difficult to identify children. Well-known algorithms for pedestrian recognition techniques include Faster R-CNN, YOLO, and SDD. In this paper, we use the Yolo algorithm to differentiate between adults and children among pedestrians. Furthermore, we confirm the change in detection results with the number of learning images by gradually increasing the number of learning images, and propose an improvement method to improve the detection accuracy of pedestrians.