Loading...

Skeleton Detection Base on Millimeter Wave Images by using Deep Neural Network

Morabbi Pazoki, Mehran | 2024

2 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56997 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Shabani, Mahdi; Kavehvash, Zahra
  7. Abstract:
  8. In response to the threat of terrorism, people are checked at security checkpoints. Like airports, places of pilgrimage have become increasingly important. Millimeter wave imaging systems are one type of system used; these systems have the ability to detect hidden items. Based on this technology, cylindrical and Cartesian body scanning devices have been commercialized. In cylindrical imaging, all aspects of the body are examined, and the output of this type of imaging is a sequence of images showing different views of a person. These body scanners use object detection neural networks to speed up the process of examining people. Also, in order to protect the privacy of individuals, their millimeter wave images should not be displayed. For this reason, manufacturing companies use a mannequin to display objects on the body. That is, the location of hidden objects in the millimeter wave image must be transferred to the mannequin. To approximate this transfer function, the method of detecting the key points of the body in different views is used. Methods based on deep neural networks have been used to estimate key points. For this purpose, after reviewing previous methods, a network used in this field was employed for millimeter wave data and considered as the base network to establish a suitable threshold for comparison with the final implemented method. The implemented method involves the key point detection network with a vision transformer. After the training of the mentioned two methods, the average accuracy of detecting the key points of the test data for the base network with a threshold of 0.2 was 75.74 percent, while for the vision transformer network, it was 85.21 percent. The performance speed of these two networks was 18 and 1600 frames per second, respectively. This superiority of the implemented network is evident from both results
  9. Keywords:
  10. Millimeter Wave ; Vision Transformer ; Cylindrical Scanning Device ; Key Points Detection ; Millimeter-Wave Imaging ; Deep Neural Networks ; Skeleton Action Recognition

 Digital Object List

 Bookmark

...see more