Table 5.

Comparison with the stand-alone access control systems based on facial image recognition.

Criteria	Saraf et al. [25]	Boka et al. [26]	Bakshsi et al. [27]	Sagar et al. [28]	Sajjad et al. [29]	Proposed Method
Embedded H/W	Raspberry Pi	Raspberry Pi with the Intel Movidius Neural Compute Stick	Arduino Uno	Renesas board (RL78 G13)	Raspberry Pi 3 model B, ARM Cortex-A53, 1.2 GHz processor, video Core IV GPU	Cortex dual core CPU, 2.50 DMIPS/MHz per core
Camera	Single camera	Two cameras	Single camera	Single camera	Raspberry Pi camera	Single camera
Face detection	Haar cascade method	Haar-feature based approach	N/A	PCA, LDA, histogram equalization	Viola Jones Method	AdaBoost with LBP
Facial feature	LBP histogram	FaceNet	Global feature with principal component analysis	PCA, LDA, histogram equalization	Bag of words with oriented FAST and rotated BRIEF	Gaussian derivative filter-LBP histogram
Face identification	N/A	k-NN	Euclidean Distance	N/A	Support vector machine	Histogram intersection
Execution time	N/A	N/A	N/A	10 sec for authentication	9 s for first 10 matches	190 ms per image
Overall accuracy	N/A	99.60% accuracy on LFW dataset	N/A	90.00% accuracy on own dataset	91.33% average accuracy on Face-95 dataset	95.00% true accept rate on own video stream dataset