Human Detection and Action Recognition for Search and Rescue in Disasters Using YOLOv3 Algorithm

Table 1

Existing papers on NDMM and SAR.

S. no	Author (s)	Objectives	Model	Data source	Dataset	Accuracy	Precision	Recall	F-measure
1	Bagloee S. A. et al. (2019)	To change the orientation of the highways to help and simplify emergency evacuation (called contraflow, a traffic control measure)	ML-O, SVM	Winnipeg traffic (keras)	700+	NA	NA	NA	NA
2	Brett W. Robertson et al. (2019)	To compare the results of manual coding of photographs sent on Twitter during a natural catastrophe with the development of deep learning training sets	VGG-16, MLPs	Random “spritzer” sample of 1% from Twitter directly from the Twitter and Youtube APIs	700+	64	NA	NA	NA
3	Sohaiba Iqbal et al. (2018)	To assess the efficacy of available natural catastrophe management solutions	PRISM model checker, statistical model checker	No public data sets available	800+	Fairly accurate result (written in paper)	NA	N.A	N.A
4	Sina Nayeri et al. (2018)	To reduce the total of relief operations’ weighted completion times	Multiobjective mix integer nonlinear programming model (MOMIN LP)	No public data sets available	1000+	89.4%	0.76	0.72	0.92
5	Marius Cioca et al. (2009)	To present the outcomes and solutions achieved in the event of natural disasters	Decision making	Case study	Not specified	N.A	N.A	N.A	N.A
6	Puad Siswahyudi et al. (2 019)	To assess the quality of the of their proposed framework	Conceptual model	Datasets are not publicly available	—	—	—	—	—
7	Muhammad Aqib et al. (2018)	To predict urban traffic behavior using deep learning algorithms	GPUs, cloud computing, deep learning techniques	Dataset taken from UK department of transport https://data.gov.uk/	1000+	93–95	85	96	88
8	M. Alfi et al. (2019)	To study on development of NDMM teaching materials	Data mining techniques	Data collected through questionnaires and tests	1200+	86.2%	0.71	0.88	0.94
9	K. Banujan et al. (20 18)	To analyzing post disaster state	Data mining techniques and geotagging	News API and Twitter API	1000 +	More precisely: 24%, moderate accuracy: 15 per cent, less accurate: 13 per cent	85% for twitter api and 92% for news api	N.A	N.A
10	Muhammad Saqib et al. (2018)	To analyze real-time drone surveillance	R-CNN	Real time data	1970 frames	72–78	0.82	NA	NA
11	Jaume Rossello et al. (2020)	To determine the effects of natural disasters	Gravity model	United nations world tourism organization (UNWTO, 2015) datasets	Data’ s of 171 countries	82%	NA	NA	NA
12	Shohei Matsuura and Khamarrul Azahari Razak (2019)	To explore how collaboration between science and technology groups and other stakeholders in the Asia-Pacific region could contribute to domestic and local disaster risk reduction	TDA, DRR	Case studies	700	NA	NA	NA	NA
13	Júlio Mendonça et al. (2019)	To use IT systems for disaster recovery solutions	BLE technology, low cost localization algorithm	Case studies	49	NA	NA	NA	NA
14	Yu-Wei Kao et al. (2019)	To present the IoT architecture for SAR and DR missions and validate it with the development of a proof of concept prototype	Hybrid for simulation and hardware looping	Live simulations	700	NA	NA	NA	NA
15	Khan Muhammad et al. (2018)	To showcase the IoT architecture for SAR and DR (disaster relief) missions and validate it with the development of a proof of concept prototype	Fine-tuned CNN, IoMT (internet of multimedia things)	Datasets taken from other literature	68,475	94–95	82	98	89
16	Shunichi Koshimura et al. (2020)	Case study of 2011 Tohoku earthquake tsunami using U-Net CNN	U-Net CNN	Case study including WorldView-2	31,262 samples	70–71	NA	NA	76
17	Baojun Zhang (2012)	To investigate emergency relief preparations in the event of a natural disaster	Conceptual model	No public data sets available	600+	NA	NA	N.A	N.A
18	Ugur Alganci et al. (2020)	To compare the evaluation of advanced CNN object detection models to determine aeroplanes from satellite images	YOLOv3, SSD and faster R-CNN	DOTA dataset	1631 images	—	Faster R-CNN (best performed)–0.98	0.92	0.94
19	Eleftherios Lygouras et al. (2019)	To detect and rescue of humans using UAV technologies and real-time computational vision with deep learning techniques	CNN with Tiny YOLOv3	Own dataset	4500 images	55%	—	70%	—
20	Khan Muhammad et al. (2017)	To identify early stage fires during surveillance	CNN	Own dataset (31 videos)	68,457 images	94.39	0.82	0.98	0.89
21	Satoshi Togawa and Kazuhide Kanenishi (2016)	To build a disaster recovery framework as against earthquake, tsunami and massive flooding	Prototype system with IaaS architecture	—	—	—	—	—	—
22	Balmukund Mishra et al. (2020)	To detect the human action for search and rescue	CNN, faster R-CNN, R-CNN	https://www.leadingindia.ai/data-set	8000	—	—	—	—
23	Muhammad Imran et al. (2014)	To perform automated classification of microblogging communications related to crises	Artificial intelligence for disaster response (AIDR)	Twitter data	—	—	—	—	—
24	Muhammad Imran et al. (2013)	To retrieve data nuggets from microblogging messages at a time of disaster	Automatic classifier	Twitter data	206,764 unique tweets	—	—	—	—
25	Saeed Hamood Alsamhi et al. (2020)	To assess the network performance of UAV-supported intelligent edge computing	OPNET	—	—	—	—	—
26	Edward J. Glantz et al. (2020)	To provide a short overview of how UAV capabilities used in disaster management	—	—	—		—	—	—
27	Lu Tan et al. (2021)	To identify pill for ensuring the safe medication delivery to patients	RetinaNet, single Shot multibox detector (SSD), and you only look once v3 (YOLOv3)	Own dataset	51,310 images	—	—	—	—
28	Muhammad Shahir Hakimy Salem et al. (2021)	To examine the use of UAV-based human detection and technology in search and rescue operations during natural disasters	—	—	—	—	—	—	—
29	Amina Khan et al. (2022)	To focus the combined role that WSN, IoT, and UAV systems could play in both natural and man-made DM	—	—	—	—	—	—	—
30	Zhang, N et al. (2022)	To offer a method for creating training visuals that were harmoniously composited	YOLOv5l	COCO dataset	—	—	44.9%	—	—
31	Kaushlendra Sharma et al. (2022)	To introduce the use of robots for the initial investigation of the disaster site. The robots toured the area and used the video stream (with audio) they had recorded to locate the human survivors	YOLOv3 (human detection) CNN (speech)	COCO dataset (human detection) GTZAN, scheirer-slaney, and MUSAN (speech)	—	−95.83%	−70.2%	—	−0.9186
32	Saeed Hamood Alsamhi et al. (2021)	To concentrate on network performance for effective disaster management collaboration of drone edge intelligence and smart wearable devices	OPNET 14.5 simulator	—	—	—	—	—	—