S.

No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

1 From Detection to ResearchGate researchgate.net/pu Delayed medical To introduce an Hybrid AI Proposes a

Action: A (Preprint) blication/38659905 care and automated highway Framework: Uses complete end-to-
Multimodal AI 1 subsequent safety management YOLOv11s for real- end pipeline from
Framework for crashes due to late framework that time visual visual detection to
Traffic Incident or non-existent integrates computer detection of automated,
Response alerts to emergency vision and NLP for accidents and fires. context-aware,
responders and real-time A Vision-Language multi-agency alert
nearby vehicles. monitoring, Model generation. The
Lack of integrated analysis, and (Moondream2) system uses a
systems that handle reporting of traffic generates textual sequence of
detection, incidents, descriptions of the specialized AI
contextual coordinating a scene. A Large models to translate
understanding, and multi-agency Language Model visual data into
actionable response. (GPT-4 Turbo) actionable
response processes this text intelligence,
generation. to create concise, demonstrating a
actionable robust framework
emergency alerts for coordinating
and response police, fire, and
recommendations. ambulance
services.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

2 Development of a ResearchGate researchgate.net/pu Traditional To develop an AI- Decision Support The system
Decision Support (Preprint) blication/39009022 emergency driven Decision System (DSS): significantly
System for 8 response systems Support System Integrates real-time improves response
Ambulance struggle with real- (DSS) to enhance geospatial data times and resource
Dispatch and time decision- the efficiency of (GPS, traffic), utilization by moving
Resource Allocation making in large- ambulance machine learning beyond simple
in Large-Scale scale incidents due dispatch and for demand "dispatch-to-
Emergencies to dynamic resource allocation prediction, and nearest" logic. It
variables like traffic, during mass optimization optimizes the entire
hospital capacity, casualty incidents algorithms patient distribution
and patient by providing data- (Dijkstra's, queuing process, preventing
prioritization, driven theory, linear ER saturation and
leading to recommendations. programming) for ensuring casualties
inefficiencies. route optimization are transported to
and hospital load the most
balancing based on appropriate and
live hospital available facilities,
capacity data. thereby improving
overall survival
rates.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

3 Optimized ResearchGate researchgate.net/pu Traditional To optimize Graph Theory & Demonstrates a 15-
Ambulance Routing (Preprint) blication/38998924 ambulance routing ambulance routing Machine Learning: 25% reduction in
and Dispatch using 1 systems fail to and dispatch by Road networks are ambulance
Graph Theory and adapt to dynamic, integrating graph modeled as graphs. response times in
Machine Learning real-time traffic theory and machine ML models (RNNs, simulations. The
for Dynamic Traffic conditions, leading learning to create a LSTMs) predict real- system's key
Conditions to critical delays in system that adapts time traffic innovation is its
reaching accident to dynamic traffic congestion, which ability to find the
scenes. conditions, thereby is used to truly fastest route in
reducing response dynamically update real-time by
times. the edge weights predicting
(travel times) of the congestion, rather
graph. than relying on
Reinforcement static, distance-
learning is used to based shortest-path
learn optimal calculations.
dispatch strategies
over time.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

4 IoT and Computer ResearchGate researchgate.net/pu Modern urban To design a smart Hybrid Edge-Cloud Proposes a
Vision for Urban (Preprint) blication/38904681 traffic presents urban traffic Architecture: A comprehensive,
Traffic Management 5 multifaceted management distributed network scalable framework
challenges of system that of IoT sensors where different
congestion, safety, optimizes traffic (Smart Cameras, sensor modalities
and regulatory flow, improves LiDAR, Radar, complement each
adherence that safety, and provides Environmental other. LiDAR and
traditional systems real-time Sensors) captures radar provide
cannot adequately intelligence by real-time data. Edge robustness in
manage. synergistically nodes perform adverse conditions
integrating IoT and localized analysis where cameras fail.
computer vision. for immediate The edge-cloud
responses. architecture
Computer vision balances low-
models (YOLOv8, latency local
Deep SORT) and ML response with
models powerful centralized
(Reinforcement analytics and model
Learning for signal training.
control) are used for
analysis.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

5 AI and IOT Based International researchgate.net/pu Existing accident To develop a Hybrid IoT and AI: This paper directly
Road Accident Journal of Scientific blication/36918482 detection systems, cognitive, agent- An IoT module with addresses the
Detection and Research in 5 particularly those based collision a force sensor critical problem of
Reporting System Engineering and relying only on IoT detection and smart detects a potential false positives. The
Management sensors, suffer from alarm system that impact. This trigger two-phase
a high rate of false uses a hybrid activates a camera approach (IoT for
alarms, leading to approach to reduce connected to a initial trigger, AI for
wasted emergency the false detection Raspberry Pi. A verification)
resources. rate and provide deep learning provides a practical
timely, verified module in the cloud solution to enhance
alerts. uses pre-trained system reliability,
models (ResNet-50, ensuring that
InceptionResNetV2) emergency services
to analyze the are dispatched only
captured video and for confirmed
verify if a genuine incidents.
accident occurred
before sending
alerts.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

6 ACCIDENT Journal of Global General problem of To review the Multi-Sensor IoT & Provides a broad
DETECTION AND Research in improving road trends, CV: Reviews overview of the
REPORTING Mathematical safety and reducing technologies, and systems using a field, summarizing
SYSTEMS IN IOT- Archives(JGRMA) fatalities from traffic challenges in IoT- suite of sensors the key
ENABLED accidents. enabled accident (accelerometers, technologies
VEHICLES: TRENDS, detection and GPS, pressure involved. It
TECHNOLOGIES, reporting systems. sensors, highlights persistent
AND CHALLENGES microphones) for challenges such as
detection. Data is sensor accuracy,
processed by false alarms,
microcontrollers connectivity, and
and smartphones. data privacy, while
ML is used for pointing to
severity emerging trends like
classification, and federated learning
GSM/5G for and blockchain for
communication. improved
Also discusses robustness.
OpenCV for visual
detection.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

7 Advancing Dynamic Mathematics doi.org/10.3390/ma Traditional Vehicle To establish a Deep Addresses the
Emergency Route th13020127 Routing Problems scientific path Reinforcement complex, dynamic
Optimization with a (VRPs) and existing strategy for rescue Learning (DRL): nature of large-
Composite Network heuristic algorithms teams by Uses an actor-critic scale disaster
Deep are not suitable for developing a DRL- framework response. The
Reinforcement emergency logistics based method that combined with proposed DRL
Learning Model where the number can handle dynamic attention model is more
and location of changes, minimize mechanisms, adaptable than
rescue points can costs, and improve pointer networks, traditional
change scheduling and LSTM neural optimization
dynamically. efficiency. networks. A algorithms, capable
composite network of finding effective
model combines routing solutions
multiple critic even when
networks to improve conditions change
performance and unpredictably
generalization. during the response
operation.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

8 AI-based American Journal of researchgate.net/pu Lack of a To systematically Systematic This comprehensive

Emergency Advanced blication/38968381 comprehensive review and Literature Review: review confirms that
Response Systems: Technology and 7 overview of how AI investigate the Analyzes 424 AI has significantly
A Systematic Engineering is being integrated integration of AI studies covering AI improved predictive
Literature Review on Solutions across various (ML, DL, CV, IoT) in applications in accuracy,
Smart Infrastructure domains of optimizing disaster automated hazard
Safety emergency emergency management, real- detection, and
response to response time incident optimized resource
enhance smart mechanisms across detection, allocation across all
infrastructure multiple sectors, healthcare reviewed domains.
safety. including disaster response, industrial It validates the
management and hazards, and transformative role
intelligent traffic intelligent traffic of AI in modern
control. control. emergency
management,
leading to faster
response times and
enhanced
situational
awareness.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

9 Optimizing NextBillion.ai Blog nextbillion.ai/blog/o Traditional To enhance EMS AI-driven Summarizes the
Ambulance ptimizing- ambulance efficiency by using Techniques: practical
Dispatch Routes ambulance- dispatch relies on AI to optimize Utilizes real-time application of AI in
using AI dispatch-routes- static routes and ambulance route optimization EMS logistics. AI
using-ai manual decision- dispatch routes, (integrating GPS and improves response
making, leading to minimize response traffic data), by moving from
delays from traffic times, and improve predictive analytics static to dynamic
congestion and patient care for demand decision-making,
suboptimal outcomes. forecasting, using predictive
resource utilization. automated models to
dispatching anticipate demand
algorithms, and and real-time data
integration with to navigate around
IoT/Smart City obstacles, ensuring
infrastructure (e.g., faster and more
smart traffic lights). efficient patient
transport.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

10 A CNN-Based International ijarsct.co.in/Paper1 The increasing To develop and CNN Architecture: Demonstrates a
Framework for Journal of Advanced 8486.pdf reliance on video deploy a CNN Uses a pre-trained practical end-to-
Video Analysis and Research in surveillance model for MobileNetV2 base end workflow from
Accident Detection Science, necessitates automatic accident for feature model development
Communication efficient and detection, making it extraction, followed to deployment. The
and Technology accurate accessible via a by a custom use of a lightweight
automated user-friendly classification head. pre-trained model
methods for frontend and The model is and conversion to
identifying suitable for trained on grayscale TFLite makes the
accidents in CCTV resource- video frames and solution viable for
footage to improve constrained converted to real-world
road safety. devices. TensorFlow Lite applications on
(TFLite) for edge devices,
deployment. A achieving an
Streamlit impressive 92%
application accuracy.
provides the user
interface.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

11 Predictive Analytics RIT Scholar Works repository.rit.edu/cg Traditional traffic To use machine Machine Learning The study found
of Road Traffic (Thesis) i/viewcontent.cgi?ar safety management learning techniques Classifiers: that the Random
Incidents, A ticle=13037 is reactive. There is to forecast the Compares the Forest model
Machine Learning a critical need for severity of traffic performance of performed best with
Approach reliable predictive incidents by Random Forest, ~84% accuracy. It
technologies to identifying crucial Decision Tree, and highlights the
anticipate contributing factors K-Nearest Neighbor feasibility of using
accidents, enabling from historical (KNN) models on historical data to
proactive safety data. the US-Accidents predict future
measures and more Kaggle dataset to accident severity,
efficient resource predict accident providing actionable
deployment. severity (a multi- intelligence for
class problem). traffic authorities to
implement
preventive
measures in high-
risk areas.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

12 Real-time Vehicle International ijsdr.org/papers/IJS Existing accident To propose and Object Detection The system
Collision Detection Journal of Scientific DR2307044.pdf detection systems develop a real-time (YOLOv3): A pre- achieves high
using YOLOv3 and Development and (sensor or vision- accident detection trained YOLOv3 precision for
Bounding Box Research based) have and alerting system model (on COCO specific accident
Methodology limitations in using YOLOv3 to dataset) is fine- classes (94% for
reliability, accuracy, specifically detect tuned on a custom rollover, 93% for
and the ability to vehicle rollovers, dataset of accident rear-end, 92% for
detect specific rear-end collisions, images. The system head-on) with very
types of collisions in and head-on processes video low latency (0.03s
real-time. collisions. frames to detect per frame on a
and classify these GPU),
specific accident demonstrating the
types. effectiveness of
fine-tuned object
detection models
for real-time,
granular accident
analysis.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

13 HIGHWAY International ijctjournal.org/wp- Traditional incident To develop a reliable Object Detection Provides a blueprint
INCIDENT ALERT Journal of Computer content/uploads/20 detection methods and efficient (YOLOv8) & API for a complete, low-
SYSTEM BY USING & Technology 25/03/HIGHWAY- are often delayed, solution for timely Integration: Uses cost alert system. It
MACHINE INCIDENT-ALERT- increasing the risk detection, reporting, YOLOv8 for real- integrates a state-
LEARNING SYSTEM-BY.pdf of secondary and communication time collision of-the-art detection
collisions and of highway incidents detection from model (YOLOv8)
hindering timely using modern CCTV feeds. Upon with a commercial
emergency technology. detection, it triggers communication API
response. visual, audio, and to create a practical,
SMS alerts using the end-to-end solution
Twilio API to notify that bridges the gap
emergency services between detection
and contacts with and notification.
location and
incident details.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

14 An Automatic 2021 IEEE 6th ujcontent.uj.ac.za/vi High road accident To design a robust, IoT & RF This paper uniquely
Vehicle Accident International Forum ew/pdfCoverPage fatalities in low-cost, automatic Communication: combines accident
Detection and Alert on Research and developing accident detection An in-vehicle IoT detection with
System with a Smart Technology for countries are due to and alert system unit with an MPU- active traffic
Traffic Light Society and Industry delays in reporting that also includes a 6050 accelerometer management for
Management (RTSI) and emergency mechanism for detects crashes and emergency
System for an vehicles being stuck giving ambulances sends GPS location responders. The
Ambulance in traffic. priority at traffic via a GSM module. smart traffic light
lights. Ambulances are component directly
equipped with an RF addresses the "last
transmitter that mile" problem,
communicates with demonstrating a
RF receivers at holistic approach to
traffic junctions to minimizing total
turn the light green response time from
in their direction of detection to arrival
travel. at the scene.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

15 Anomaly Detection Sensors doi.org/10.3390/s22 Manual monitoring To propose a CNN & Rolling The paper
from Surveillance 176563 of vast numbers of methodology for Prediction: A CNN addresses the
Videos: A Review on traffic surveillance automatically model (pre-trained practical challenge
Deep Learning- screens is detecting traffic ResNet50) is trained of applying frame-
Based Approaches inefficient and accidents to classify individual based classifiers to
prone to human (anomalies) in video frames as video streams. The
error, leading to surveillance videos accident or non- rolling average
missed accident using a CNN and a accident. A rolling technique is a
events. rolling prediction prediction average simple but effective
algorithm to algorithm is then post-processing
improve accuracy. applied to the step that improves
sequence of the temporal
classifications to consistency and
smooth out reliability of the final
predictions and detection output,
reduce "label achieving 82%
flickering" or accuracy.
momentary
misclassifications.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

16 A Hybrid Method for Mathematics doi.org/10.3390/ma Existing traffic To propose a hybrid Hybrid ViT & CNN: This hybrid model
Traffic Accident th11041050 accident model that fuses A Trajectory achieves ~97%
Classification classification features from both Generator uses accuracy by
Based on CCTV methods struggle video frames and YOLOv5 and Deep combining spatial
Frames with complex object trajectories SORT to detect information from
scenes, have poor to improve objects and track the frame (via ViT's
generalization, and classification their paths. A Traffic attention
often cannot accuracy and help Accident Classifier mechanism) and
determine the locate the accident then uses a Vision temporal/motion
precise location of area. Transformer (ViT) to information from
an accident within a analyze the raw trajectories (via
frame. frame and a CNN to CNN). This fusion
analyze the provides a richer set
trajectory map. The of features, leading
features from both to more robust and
are fused for a final accurate
classification. detection.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

17 Real-time fire Sensors doi.org/10.3390/s23 Traditional sensor- To present a fire Object Detection Achieves 93%
detection for 167078 based fire detection detection technique (YOLOv7): The accuracy,
maritime systems (smoke, based on YOLOv7 to YOLOv7 model is demonstrating that
environment using thermal) are often provide timely and trained on a custom vision-based
YOLOv7 slow to react and accurate detection dataset of fire systems can be
require human in challenging images, including significantly faster
verification, environments, day and night and more direct
allowing fires to specifically for scenarios, to detect than traditional
grow to a dangerous maritime safety. the visual presence sensors. The use of
extent before an of flames in real- a state-of-the-art
alarm is confirmed. time. object detection
model allows for
the detection of
small or incipient
fires, enabling a
much quicker
response.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

18 Vision based real- Multimedia Tools pmc.ncbi.nlm.nih.g Non-vision sensors To develop a cost- Object Detection & The proposed
time traffic and Applications ov/articles/PMC991 (LIDAR, RADAR, effective, vision- Custom Tracking: centroid-based
monitoring system 1956/ etc.) are expensive based real-time Uses a customized tracker is shown to
for intersections to install and traffic flow YOLOv3 for vehicle be 9.18% more
maintain. Vision- monitoring system detection. A novel accurate and
based systems face for intersections centroid-based significantly faster
challenges with that is robust to tracking algorithm is (up to 30 fps) than a
occlusion, weather, common visual proposed, which standard Kalman
and camera challenges. tracks the shift in an filter approach,
placement. object's bounding especially in dense
box centroid across traffic. This
frames and uses highlights the
Euclidean distance importance of
for data domain-specific
association. tracking algorithms
for robust
performance.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

19 IoT-based Vehicle World Journal of Existing advanced To reduce the threat IoT Sensor Suite & This system moves
Accident-Avoidance Advanced Research safety features are of collisions Automated beyond simple
System and Reviews often costly and through real-time Braking: Uses detection to active
exclusive to high- sensing, ultrasonic sensors collision avoidance.
end vehicles. There processing, and to detect obstacles The inclusion of an
is a need for a low- automatic alarm and an MPU6050 automated braking
cost system that generation, offering (accelerometer/gyro mechanism
can provide real- a safer and more scope) to monitor represents a step
time alerts and efficient vehicle motion. An towards Advanced
automated transportation ESP32 Driver-Assistance
intervention. environment. microcontroller Systems (ADAS) in a
processes the data. low-cost,
If a hazard is accessible
detected and the package.
driver does not
respond to a buzzer
alert, an automatic
braking system is
activated.
S.No Title of the paper Title of the Journal Reference with DOI Problem or gap Objective of the Focus of the paper Summary
addressed Research (Discuss methods
Mentioned used)

20 Vision-based traffic arXiv (Preprint) arxiv.org/pdf/2505.0 Lack of a structured To provide a Systematic This comprehensive
accident 7611 overview of the foundational Literature Review: survey identifies key
anticipation (Vision- rapidly evolving field reference by Reviews 147 challenges like data
TAA): A survey on of predicting traffic synthesizing papers, categorizing scarcity, limited
deep learning accidents from existing deep learning generalization, and
approaches video feeds before advancements, methods into four real-time
they happen. categorizing approaches: performance
methodologies, and Image/Video constraints. It
identifying critical Feature-Based, highlights future
gaps and future Spatio-Temporal opportunities in
research directions Feature-Based, multimodal fusion,
in Vision-TAA. Scene self-supervised
Understanding- learning, and
Based, and Transformer
Multimodal Data architectures,
Fusion-Based. providing a
roadmap for the
future of accident
prediction
research.