A New Approach for Human Body Orientation Detection Using Camera Footage


Kartal A., Shahbazi N., SEZER E., GÜZEL M. S.

8th International Conference on Computer Science and Engineering, UBMK 2023, Burdur, Turkey, 13 - 15 September 2023, pp.46-50 identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/ubmk59864.2023.10286590
  • City: Burdur
  • Country: Turkey
  • Page Numbers: pp.46-50
  • Keywords: camera footage, computer vision, object detection, orientation detection, person detection
  • Hacettepe University Affiliated: Yes

Abstract

This study contributes to the field of orientation detection from camera footage, showcasing the advancements in computer vision techniques and their practical applications. The insights gained from this study can be leveraged in various domains, including surveillance, security, activity recognition, and pose estimation, fostering the development of more efficient and accurate orientation detection systems. Impressive results are obtained, with high AUROC values achieved for classifying different orientations: 0.91 for 'right,' 0.91 for 'left,' 0.95 for 'fore,' and 0.93 for 'back.' These results underline the models' ability to accurately estimate human body orientations as fore, back, right, and left from camera footages with a high degree of accuracy.