HOG 알고리즘과 CNN을 이용한 객체 검출 시스템에 관한 연구

• Published in: (사)디지털산업정보학회 논문지, 20(3) Vol. 20; no. 3; pp. 13 - 23
• Main Authors: 박병준, 김현식, Park Byungjoon, Kim Hyunsik
• Format: Journal Article
• Language: Korean
• Published: (사)디지털산업정보학회 01.09.2024
• Subjects: 컴퓨터학; Histogram of Oriented Gradients(HOG); Image Processing; Objects Tracking; Convolution Neural Network(CNN)
• ISSN: 1738-6667; 2713-9018

For the purpose of predicting credit card customer churn accurately through data analysis Detecting and tracking objects in continuous video is essential in self-driving cars, security and surveillance systems, sports analytics, medical image processing, and more. Correlation tracking methods such as Normalized Cross Correlation(NCC) and Sum of Absolute Differences(SAD) are used as an effective way to measure the similarity between two images. NCC, a representative correlation tracking method, has been useful in real-time environments because it is relatively simple to compute and effective. However, correlation tracking methods are sensitive to rotation and size changes of objects, making them difficult to apply to real-time changing videos. To overcome these limitations, this paper proposes an object tracking method using the Histogram of Oriented Gradients(HOG) feature to effectively obtain object data and the Convolution Neural Network(CNN) algorithm. By using the two algorithms, the shape and structure of the object can be effectively represented and learned, resulting in more reliable and accurate object tracking. In this paper, the performance of the proposed method is verified through experiments and its superiority is demonstrated.