Auditory Speech Based Alerting System for Detecting Dummy Number Plate via Video Processing Data sets

Table 1

Comparison of different versions of YOLO [5].

YOLO v3	YOLO v2	YOLO	Detector
	—	—	Proposal
Darknet	Darknet	GoogleNet	Backbone
Variable	Fixed	Fixed	Input image
	<50	<25 VGG	Speed
CVPR18	CVPR17	CVPR16	Published in
Sigmoid instead of a softmax	SGD	SGD	Optimization
Cross-entropy error terms + bounding box regression + object confidence (object confidence and class predictions in YOLO v3 are predicted through logistic)	Class sum square error loss + bounding box regression + object confidence + background confidence	Class sum square error loss + bounding box regression + object confidence + background confidence	Loss function
	Yes	Yes	End-to-end train
C and python	C	C	Language
Darknet-53	Darknet-19	Darknet	Deep learning platform
Achieves good performance for small objects as well as with more speed	Achieve high accuracy and high speed propose a faster darknet 19; improved the speed and accuracy by using several existing strategies; YOLO 9000 can detect over 9000 object categories in real-time limitations: struggling in detecting small objects	First unified detector framework (elegant and efficient) exclude RP method completely, faster than previously proposed detectors. YOLO and fast YOLO run at 45 and 155 FPS, respectively limitations: have difficulty to localize tiny objects. Dramatics accuracy falls as compared to the state of art	Merits and limitations