JISE


  [ 1 ] [ 2 ] [ 3 ] [ 4 ] [ 5 ] [ 6 ] [ 7 ] [ 8 ] [ 9 ] [ 10 ] [ 11 ] [ 12 ] [ 13 ]


Journal of Information Science and Engineering, Vol. 37 No. 3, pp. 593-604


Statistical Multiframe Methodology with Agnostic Thresholding for Attendance Marking System


KUAN HENG LEE, SANJAY V. ADDICAM, ILYA KRYLOV, SERGEI NOSOV,
MEE SIM LAI, ZHAN QIANG LEE AND CHUNG SHIEN CHAI
Intel Microelectronics Sdn. Bhd.
IOTG Retail Banking Hospitality Education
Bayan Lepas, Penang, 11900 Malaysia
E-mail: {kuan.heng.lee; addicam.v.sanjay; ilya.krylov; sergei.nosov; mee.sim.lai;
zhan.qiang.lee; chung.shien.chai}@intel.com


Attendance marking is a burdensome and time-consuming task for every school teaching staff to perform manually in the classroom. It becomes very attractive if this attendance marking process can be automated through a facial recognition system. Although facial recognition works well under constrained environment, identifying each student in a dynamic classroom environment remains a challenge especially the students are in uncooperative manner. Conventional frame-based accuracy metric cannot reflect the true outcome of the attendance as it varies drastically over frames, due to the large variations of scales, poses and occlusions in the classroom environment. In this paper, a statistical methodology based on multiframe was proposed to improve the attendance marking accuracy after a convergence time. This methodology was combined with the mean thresholding scheme to achieve the same accuracy as full inference rate (i.e. 30 FPS) with a lower inference rate (i.e. 3 FPS). This drives away the need to invest an expensive hardware to maintain the same accuracy with a higher inference rate.


Keywords: artificial intelligence (AI), face detection (FD), face recognition (FR), interpupillary distance (IPD), sliding window filtering, frame per second (FPS), false positive (FP), false negative (FN), thresholding

  Retrieve PDF document (JISE_202103_06.pdf)