Computing (2022) 104:1203–1217

https://doi.org/10.1007/s00607-021-01038-1

REGULAR PAPER

Machine learning driven intelligent and self adaptive

system for traffic management in smart cities

Hameed Khan1 · Kamal K. Kushwah1 · Muni Raj Maurya2,3 · Saurabh Singh1 ·

Prashant Jha1 · Sujeet K. Mahobia1 · Sanjay Soni1 · Subham Sahu1 ·
Kishor Kumar Sadasivuni2

Received: 13 July 2021 / Accepted: 25 November 2021 / Published online: 16 January 2022
© The Author(s), under exclusive licence to Springer-Verlag GmbH Austria, part of Springer Nature 2021

Abstract
Traffic congestion is becoming a serious problem with the large number of vehicle
on the roads. In the traditional traffic control system, the timing of the green light is
adjusted regardless of the average traffic rate at the junction. Many strategies have
been introduced to solve and improve vehicle management. However, in order to
handle road traffic issues, an intelligent traffic management solution is required. This
article represents a self adaptive real-time traffic light control algorithm based on the
traffic flow. We present a machine learning approach coupled with image processing to
manage the traffic clearance at the signal junction. The proposed system utilizes single
image processing via neural network and You Only Look Once (YOLOv3) framework
to establish traffic clearance at the signal. We employed YOLO architectures because
it is accurate in terms of mean average precision (mAP), interaction over union (IOU)
values and fast in object detection tasks as well. It runs significantly faster than other
detection methods with comparable performance. The average processing time of
single image was estimated to be 1.3 s. Further based on the input from YOLO we
estimated the ‘on’ time period green light for effective traffic clearance. Several real
time parameters like number of vehicles (two wheelers, four wheelers), road width
and junction crossing time are considered to estimate the ‘on’time of green light.
Moreover, we used the real traffic images to test the performance and trained the
system with different dataset. Our experiments investigation reveals that the predicted

B Kamal K. Kushwah
[email protected]
B Saurabh Singh
[email protected]
B Kishor Kumar Sadasivuni
[email protected]

1 Jabalpur Engineering College, Jabalpur, Madhya Pradesh, India

2 Center for Advanced Materials, Qatar University, Doha, Qatar
3 Mechanical and Industrial Engineering, Qatar University, Doha, Qatar

123
1204 H. Khan et al.

vehicle counts were well matched with the actual vehicle count and proposed method
apprehended an average accuracy of 81.1%. The reported strategy is self adaptive,
highly accurate, fast and has the potential to be implemented in the traffic clearance
at the junctions.

Keywords Traffic management · Machine learning YOLO · Image processing ·

Open CV

Mathematics Subject Classification 68T05 Learning and adaptive system in AI ·

68T45 Machine vision and scene understanding · 68-11 Research data for problem
pertaining to computer science

1 Introduction

Traffic congestion is a major problem in many cities. Roads are witnessing a large
number of vehicles running each day and have made the task of managing the traffic
more tedious. Extensive research works are going on to make traffic management
system more adaptive, smart, and intelligent. Cameras have been installed on the
roads and at junctions for surveillance purposes, for imposing autonomous penalty and
person identification upon violating the traffic rules. In terms of traffic management,
majority of the traffic junctions use a fixed time green light cyclic system for managing
the traffic. The incorporation of fixed cyclic time period operation in traffic light
management system imposes certain limitations and has proven to be significantly
inefficient in regulating traffic congestions. The traditional time-limited robot system
works best when the traffic flow is almost the same in all four directions. However,
during a day, we witness a situation when traffic is more from one compared to other
directions. Moreover, the traditional system lacks smart management; as a result it
makes people wait, irrespective of no vehicle from the other side. This unavoidable
waiting time sometime makes person restless, which often ends up in violation of rules
and accidents. Further, it promotes more consumption of fuel and adds pollutants in
surrounding environment. Consistent research work has been carried out to manage
traffic congestion and automate the traffic management process. The idea of making
the traffic light adaptive to real-time traffic stream is not new, and different strategies
have been reported to achieve it. Researcher proposed the design; which consists of
three basic components: a parking management center, a traffic management center,
and world facts and management core [1]. The facility depends on the functioning of
sensory networks to accumulate traffic congestion and the range of vehicles traveling
at every intersection to make a decision when estimating the length of time the traffic
light can stay green. Several other framework of the intelligent traffic management has
been proposed in this regard [2,3]. These structured frameworks offers the STMS site
visitors manipulate machine as the main module and has small modules such as video
manipulate system, traffic control system, computer control system, and peripheral
devices. The traffic control system controls heavy visitors at a predetermined time
on the road. It uses a video surveillance device to discover extra site visitors with
a video camera and when the range of automobiles on a precise road exceeds the

123
Machine learning driven intelligent and self adaptive… 1205

pre-determined number, it notifies traffic controllers with an alarm indicating that

“traffic has been reached” and prevents any other car from entering that route. The
following cars would consequently be diverted to every other certain route, which
will lead to controlling of traffic jams. This site visitors manipulate machine includes
high-quality transmission and uninterrupted conversation by sending and receiving the
appropriate signal at the proper time. Peripheral smart devices (IoT) deal with the right
configuration of the input sensors and output actuators to capture and receive events
and send remarks as needed details for managing points [4]. Similarly, the deployment
of the CCTV camera to visualize areas of the smart road covering the scope of the
situation is also an important function of this module [5].
The potential to predict the temporary emergence of the contemporary traffic state
is an essential basis for persisted use in traffic management and control. Some stud-
ies focused on Advanced Traffic Management Systems (ATMS)/Advanced Traveler
Information Systems (ATIS) routes, a community of traffic sensors that grant real-time
traffic information [6]. Other significant cost effective approach is vehicle counting
and tracking with YOLO using real-time video data; the counting method process a
traffic video captured using handheld cameras in a video-based vehicle [7]. Based on
this a real-time traffic light control algorithm has been reported which takes input as
the real-time traffic video and optimizes the green light time accordingly [8]. Very
similar methods have been proposed to use video data to make self-adaptive traffic
light and replacing a traffic police officer with an intelligent system [9,10]. In Table 1
we provide a comparative analysis of the different types of work performed by many
researchers on different platforms. One thing is common in all the work done; all traf-
fic management systems take input as video data and process the video to get insights
about the traffic condition.
In the above reported studies, usually traffic stream video is processed as a standard
input data to manage the traffic clearance [2,6–10,22]. Video is a collection of images,
thus to process a video and get the insights of traffic, all the images need to be processed
in real-time, which is a computationally expensive and time consuming. Moreover,
the continuous processing of images extracted from the video affects the durability
of the system and is expensive for long term usage. Due to these constrains there is a
high need of an intelligent traffic management that is fast, accurate and cost effective.
In this paper, a solution to above mentioned problems has been attempted with of
machine learning tool. The YOLO object detection algorithm is used to detect cars in
real-time images captured by a camera mounted on the top of the signal and allocate
green light time according to the number of vehicles. Compared to complete video
processing the proposed self adaptive utilizes a single image processing to calculate
the optimal green light time in view of optimized memory requirements [23]. The
proposed model offer self adaptive robust facility to makes traffic management more
efficient and effective.

123
1206 H. Khan et al.

Table 1 Summary of existing relevant studies and applications

Study Focus and application Approach

Makaba et al. [11] Bayesian network-based Simulation framework

framework
Khan et al. [12] Aerial vehicle-based traffic Empirical study
analysis
Takano et al. [13] High-resolution image data Numerical simulation
collection
Fielbaum [14] Impact of traffic congestion Numerical simulations
Lykov and Asakura [15] Tensor-based abnormal Numerical simulation
pattern detection for traffic
Contreras and Gamess [16] Algorithm to count vehicles Numerous simulations
at a traffic signal
Hu et al. [17] Recognition of traffic density Traffic Simulator
Peque et al. [18] Adaptive learning algorithms Mathematical formulation
for dynamic traffic
Hamidi and Kamankesh [19] Intelligent traffic Multi-agent system
management system
García et al. [20] Intelligent method for traffic IOCA-PSO method
light scheduling
Jain [21] Automatic traffic signal Membership function based
controller

2 Proposed methodology

2.1 System architecture

The system requires four cameras, each at the top of the road that meets on a junction
(see Fig. 1). These cameras are used to send the image to an embedded controller.
After that, when the controller receives the real-time image, it will count the number
of vehicles present in the image and light up the green signal according to the count
of vehicles. Most of the traffic junctions have such kind of cameras already installed
on top of the signals. We need to connect them with a controller. The controller would
be able to perform the operations listed below:
1. Instruct the camera to click the picture of the traffic situation.
2. Process the input image from the camera and count the number of vehicles in the
image.
3. Calculate the time duration, as to how long the green light should remain in on
state based on the vehicle count.
4. Light up the green light of the corresponding signal and the red light of the other
signals.
As shown in Fig. 2, cameras are covering only one side of the road. This site is the
driving side from where the vehicles are arriving on the signal. The cameras have to
be installed such that vehicles should not overlap each other in the image; otherwise, it
can lead us to false detection results. Figure 3 shows the different type of images taken

123
Machine learning driven intelligent and self adaptive… 1207

Fig. 1 The side view of the camera placement

Fig. 2 Top view of the camera placement

on the traffic signal. From figure it’s clear that images on the left side are not suitable
for processing, as the percentage of vehicle overlapping is significantly high. On the
other hand, the two images on the right portion of Fig. 3, gives a clear distribution
of the vehicle with minimum overlapping. Thus, these types of images will be very
helpful for the present proposed method.

2.2 System architecture

In this work, OpenCV is used for counting the vehicles from the input image. OpenCV
is an image processing library written in C++ and Python. OpenCV [5] provides mod-
ules to run object detection with various methods, and here in this work, OpenCV.cdn
[24] module is used to count the number of vehicles. The input image is provided to
the YOLO v3 model through OpenCV [25,26]. YOLO is a smart convolution neural
network (CNN) for object detection in real-time. The algorithm applies a single neu-

123
1208 H. Khan et al.

Fig. 3 The sample images for training purposes

ral network to the full image, and then divides the image into regions and predicts
bounding boxes and probabilities for each region. These bounding boxes are weighted
by the predicted probabilities. It returns a list of bounding box coordinates for each
object detected in the image. The number of vehicles would be the number of objects
in the returned list. Majority of the study consider equal time lane clearance time for
two-wheelers (TW) and four-wheelers (FW) vehicles (see Fig. 4) [27,28]. However,
considering the real time scenario the FW take more time than TW vehicles in starting
and leaving the signal. Thus, it’s important to consider the time taken to leave the sig-
nal junction also. In this work, we have considered that TW vehicles and FW vehicles
contribute differently to the time required to clear the lane. Based on the number of
vehicles and their respective time to clear the lane the the actual time period is esti-
mated and finally the controller allocate the on time green light which is represented
as ‘t’. The time ‘t’ is determined by Eq. (1).

t = T W /a × n + F W /b × m (1)

where TW = Number of two-wheelers/bikes in the image; FW = Number of four-

wheelers/cars in the image; n = Time to clear single two-wheeler; m = Time to clear
single four-wheeler; a = maximum no. of two-wheeler covers the whole width of the
road; b = maximum no. of four-wheeler covers the whole width of the road; x Denotes
ceil of x.

123
Machine learning driven intelligent and self adaptive… 1209

Fig. 4 TCS stages and flow of communication

Clearance time of the individual vehicle is considered by considering the time taken
by the vehicle to cross the stop line, when it is standing just before the stop line. Thus
constants n, m, are determined by the real-time observation of the traffic junction. The
measured value for n and m are 3.6 and 6 s respectively. Since the reaction time of
a driver in situations is generally greater than 1 s, thus we have rounded the value
of n to 4 s. To calculate constants a, b, we need to consider more parameters like,
road width (Wroad), width of a TW (W2), width of a FW (W4), and the gap between
any two vehicles (G). Road width may vary for different traffic junctions. Here we
are considering the common case. Wroad = 30 ± 5 feet, W2 = 2 feet, W4 = 6 feet,
G = 1 feet, then a and b are calculated using Eqs. (2) and (3), respectively.

 
Wr oad
a= (2)
W2 + G
 
Wr oad
b= (3)
W4 + G

For, road width = 25 ft.

   
25 25
a= = 8.33 = 8; b = = 3.57 = 3
2+1 6+1

For, road width = 30 ft.

   
30 30
a= = 10 = 10; b = = 4.28 = 4
2+1 6+1

For, road width = 35 ft.

   
35 35
a= = 11.67 = 11; b = = 5 = 5
2+1 6+1

123
1210 H. Khan et al.

wherex denotes floor of x

There is a huge possibility that there are too many vehicles waiting at the junction
and will take long time for them to pass, which will lead to starvation for vehicles
on the another side of road junction. Thus to overcome this need to set a maximum
threshold time and choose the minimum value between the time estimated from the
equations mentioned above. Considering this situation the actual time to calculate the
on time for green light is given by Eq. 4

tactual = min(t, M AX _G R E E N _T I M E) (4)

To calculate MAX_GREEN_TIME, we need to consider the worst case of Eq. (1)

that is when all vehicles are FW. Assuming that on any side of the junction we need
to clear maximum 20 FW and putting TW = 0, FW = 20 in Eq. (1) and considering
different road widths the estimated time is
For road width = 25 ft.

   
0 20
t= ×4+ × 6 = 42 s
8 3

For road width = 30 ft.

   
0 20
t= ×4+ × 6 = 30 s
10 4

For road width = 35 ft.

   
0 20
t= ×4+ × 6 = 24 s
11 5

For further calculation we have considered the worst-case scenario i.e. the minimum
road width of 25 ft. that correspond to the MAX_GREEN_TIME of 42 s. Figure 5
shows the behavior of Eq. (1), with respect to number of vehicles, for three different
scenarios; (1) when traffic only has TW, (2) when traffic has only FW, and (3) when
traffic has both TW and FW in equal ratio. From Fig. 5 it is observed that irrespective
of the above three situations the curve reaches a flat for MAX_GREEN_TIME of 42 s.
To further investigate on the worst-case scenario we have only considered the FW
vehicles and estimated the clearance time, as shown in Fig. 6. From the figure it can be
observed that the calculated total clearance time of 100 FW vehicles in traffic is 546 s.
The round time mentioned in the plot is the time when a vehicle get second clearance
chance after MAX_GREEN_TIME threshold time is reached. Figure 7 represent the
flow chart of the algorithm for a four way junctions clearance framework. For each of
the four ways meeting on the intersection:

123
Machine learning driven intelligent and self adaptive… 1211

Fig. 5 Behavior of formula derived to calculate time for green light

Fig. 6 Total clearance time for traffic with respect to number of vehicles in worst case

1. Take input as the image before 2 s of turning ON the green light.

2. Preprocess the image for vehicle detection.
3. Perform object detection and count the number of (a) TW (b) and FW in the image.
4. Calculate the total time t by putting values in Eq. (1).
5. Turn the green light for the total duration of time t
6. Repeat for the adjacent camera from step-1.
The extraction of features, convolution layers are there in the model to extract them.
The data input is actually an image which is provided in the form of a numpy array

123
1212 H. Khan et al.

Fig. 7 Flowchart of the process

to the model and the model outputs the objects detected in the image with the coordi-
nates of the corresponding object. YOLOv3 is the state-of-the-art model for real-time
object detection. YOLOv3 predicts 4 coordinates for each bounding box around an
object. Training is performed with sum of squared error loss. YOLOv3 predicts an
objectness score for each bounding box using logistic regression and the class proba-
bilities using independent logistic classifiers. It uses binary cross-entropy loss for the
class predictions. YOLOv3 extracts features from 3 different scales using a concept
similar to feature pyramid networks. From the darknet-53 feature extraction back-
bone, YOLOv3 adds several convolutional layers, the last of which predicts a 3-d
tensor encoding bounding box, objectness, and class predictions. In YOLOv3’s orig-
inal COCO experiments, they predict 3 bounding boxes at each scale, so the tensor
is N N [3(4 + 1 + 80)] for the 4 bounding box offsets, 1 objectness prediction, and
80 class predictions. For the second scale, YOLOv3 takes the feature map from the 2

123
Machine learning driven intelligent and self adaptive… 1213

Table 2 Vehicle detection results on testing images

Image no. Actual count Predicted Correct False positives Accuracy (%) Time (s)
count predictions (true (%)
positive)

5 42 39 38 1 92.85 1.76717
10 18 12 11 1 66.66 1.25578
15 13 12 11 1 92.3 1.26183
31 32 32 28 4 100 1.2677
45 91 49 46 3 53.84 1.25728

previous layers and up samples it by 2×. It also concatenates in a feature map from
earlier in the network, and then adds a few more convolutional layers to process the
combined feature map, and now predict a similar tensor at the second scale. This
process is repeated at the third scale. It automatically identifies and draws rectangular
bounding boxes around objects of interest at a rate of roughly 15–20+ frames per
second. YOLO networks divide the image into regions and predict bounding boxes
and probabilities for each region. For a 416 × 416 input image, YOLOv3 predicts
13 × 13 × 3 = 507 boxes for the first scale, 26 × 26 × 3 = 2028 boxes for the
second scale, and 52 × 52 × 3 = 8112 boxes for the third scale, for a total of 10,647
boxes. Non-maxima suppression and IOU thresholds are then used to cut the number
of boxes down significantly, often to a few or a couple dozen per image. There are
several alternative state-of-the-art object detection models and some of these models
are more accurate than YOLOv3 or return more specific results. However, for all of
these alternative models, the increased accuracy comes at the cost of slower. In this
paper our objective is to detect objects in real time, hence we have to set a trade-off
between speed and accuracy.

3 Results and discussion

The YOLO object detection model was trained on 42 authentic traffic images of two
main junctions of the city Jabalpur. A total of five images were kept apart from the
training set for testing the results. Table 2 illustrates the results obtained on the test
images.
From the results in Table 2, it can be concluded that YOLO gives us fair results
considering that we have a minimal dataset that consists only of 42 images. The mean
time taken for detecting the vehicles per image is 1.36 s. The time taken in object
detection mainly depends on the incorporated processing hardware. The above results
have been obtained on Intel(R) Xeon(R) CPU @ 2.00GHz processor. In Table 2, the
results obtained for image no. 45 shows quite different results from the other four
images. It is an outlier. That image was not taken at a right angle, and it has many
vehicles overlapping each other. For correct predictions, it is necessary to take an
image that shows vehicles as discrete as possible, not overlapped by any other object,
as discussed in the system architecture section and shown in Fig. 3 (right portion

123
1214 H. Khan et al.

Fig. 8 Processing time for one image

Fig. 9 Actual object count

versus predicted object count

images). The time mentioned in Table 2 is calculated programmatically by taking

the difference between the time points just after giving the input image and the time
point just after getting output. However the time considered for illuminating the green
light was estimated to be 2 s. Figure 8 summarizes the reason behind taking 2 s before
turning the green lights ON. From the results in Table 2, the meantime for vehicle
detection and counting is 1.36 s, which can go up to 1.5 s. In some cases, about 0.4 s
has taken in image preprocessing before the vehicle detection process. In the end, the
green light time calculation can take a maximum time of 0.1 s. The whole process sums
up in 1 s. Figure 9 represents the actual object count versus the predicted count. From
the figure it is evident that, the dark blue line curve (predicted) follows the orange
cure (actual) except for image number 45, reason behind such behavior has already
been described earlier. In Fig. 10, the graph summarizes the accuracy of images based
on the object predicted in the image. As it is visible, image 31 has highest accuracy
while image 45 has the lowest accuracy. Thus the average estimated accuracy of the
adopted method was 81.1%.
The processing time of five images has been depicted in Fig. 11. Here minimum
processing time is 1.25 s, whereas the maximum processing time is 1.76 s. The mean
processing time is approximately 1.3 s. From the above investigations it can be inferred
that the proposed intelligent traffic management system based on the single image
processing is self adaptive, highly accurate, fast and has the potential to be implemented
in the traffic clearance at the junctions.

123
Machine learning driven intelligent and self adaptive… 1215

Fig. 10 Accuracy of images

Fig. 11 Image processing time

4 Conclusion and future work

In summary, a traffic management system has been proposed to make the traffic lights
adaptive to real-time traffic streams with the help of machine learning driven YOLO
method. We considered the time parameters like number of vehicles (two wheelers,
four wheelers), road width and juction crossing time to estimate the ‘on’ time of green
light. We incorporated the real traffic images as input to our neural-network and trained
the system with different dataset that improved the vehicle detection. The evaluation
results showed that the proposed system achieved satisfactory performance with an
average accuracy of 81.1%. The reported method is modest in hardware requirements
and cost effective compared to the traditional traffic clearance strategies. In addition,
it does not need large scale construction or installation work. The proposed model
facilitates that the system can be improved by making it learn by itself with the help
of reinforcement learning. This can be active by providing regular feedback to the
system so that it can learn from it. In this way, the system will become more accurate
with time. The time taken for capturing the image and calculating the time can also
be reduced from 2 s with the help of a faster object detection system. Functionality to
take care of special cases can be added in the future. For example, if a traffic camera
detects an ambulance, the system will let it pass as soon as possible. Moreover the
future work will consider the test with cameras installed at higher altitude having wide

123
1216 H. Khan et al.

and panoramic coverage to cover larger section of the road that will be further improve
the accuracy of the results.

References
1. Redmon J, Farhadi A (2018) Yolov3: an incremental improvement. arXiv preprint arXiv:1804.02767
2. Rath M (2018) Smart traffic management system for traffic control using automated mechanical and
electronic devices. IOP Conf Ser Mater Sci Eng 377:012201. https://doi.org/10.1088/1757-899X/377/
1/012201
3. Shinde MS, Jagtap S (2016) Intelligent traffic management systems. A review. Int J Innov Res Sci
Technol 2:293–298
4. Komsiyah S, Desvania E (2021) Traffic lights analysis and simulation using fuzzy inference system of
mamdani on three-signaled intersections. Procedia Comput Sci 179:268–280. https://doi.org/10.1016/
j.procs.2021.01.006
5. Toh CK, Sanguesa JA, Cano JC, Martinez FJ (2020) Advances in smart roads for future smart cities.
Proc R Soc A 476(2233):20190439. https://doi.org/10.1098/rspa.2019.0439
6. Allström A, Barceló J, Ekström J, Grumert E, Gundlegård D, Rydergren C (2017) Traffic management
for smart cities. In: Angelakis V, Tragos E, Pöhls HC, Kapovits A, Bassi A (eds) Designing, developing,
and facilitating smart cities. Springer, Cham, pp 211–240. https://doi.org/10.1007/978-3-319-44924-
1_11
7. Asha CS, Narasimhadhan AV (2018) Vehicle counting for traffic management system using YOLO and
correlation filter. In: 2018 IEEE international conference on electronics, computing and communica-
tion technologies (CONECCT). IEEE, Bangalore, pp 1–6. https://doi.org/10.1109/CONECCT.2018.
8482380
8. Alpatov BA, Babayan PV, Ershov MD (2018) Vehicle detection and counting system for real-time
traffic surveillance. In: 2018 7th Mediterranean conference on embedded computing (MECO). IEEE,
Budva, pp 1–4. https://doi.org/10.1109/MECO.2018.8406017
9. Basil E, Sawant S (2017) IoT based traffic light control system using Raspberry Pi. In: 2017 Inter-
national conference on energy, communication, data analytics and soft computing (ICECDS). IEEE,
Chennai, pp 1078–1081. https://doi.org/10.1109/ICECDS.2017.8389604. https://ieeexplore.ieee.org/
document/8389604/
10. Corovic A, Ilic V, Duric S, Marijan M, Pavkovic B (2018) The real-time detection of traffic participants
using YOLO algorithm. In: 2018 26th Telecommunications forum (TELFOR). IEEE, Belgrade, pp 1–4.
https://doi.org/10.1109/TELFOR.2018.8611986
11. Makaba T, Doorsamy W, Paul BS (2021) Bayesian network-based framework for cost-implication
assessment of road traffic collisions. Int J ITS Res 19(1):240–253. https://doi.org/10.1007/s13177-
020-00242-1
12. Khan M, Ectors W, Bellemans T, Janssens D, Wets G (2018) Unmanned aerial vehicle-based traf-
fic analysis: a case study for shockwave identification and flow parameters estimation at signalized
intersections. Remote Sens 10(3):458. https://doi.org/10.3390/rs10030458
13. Takano T, Ono S, Kawasaki H, Ikeuchi K (2021) High-resolution image data collection scheme for road
sensing using wide-angle cameras on general-use vehicle criteria to include/exclude collected images
for super resolution. Int J ITS Res 19(2):299–311. https://doi.org/10.1007/s13177-020-00243-0
14. Fielbaum A (2020) Strategic public transport design using autonomous vehicles and other new tech-
nologies. Int J ITS Res 18(2):183–191. https://doi.org/10.1007/s13177-019-00190-5
15. Lykov S, Asakura Y (2020) Anomalous traffic pattern detection in large urban areas: tensor-based
approach with continuum modeling of traffic flow. Int J ITS Res 18(1):13–21. https://doi.org/10.1007/
s13177-018-0167-5
16. Contreras M, Gamess E (2020) An algorithm based on VANET technology to count vehicles stopped
at a traffic light. Int J ITS Res 18(1):122–139. https://doi.org/10.1007/s13177-019-00184-3
17. Hu H, Gao Z, Sheng Y, Zhang C, Zheng R (2019) Traffic density recognition based on image global
texture feature. Int J ITS Res 17(3):171–180. https://doi.org/10.1007/s13177-019-00187-0
18. Peque G, Miyagi T, Kurauchi F (2018) Adaptive learning algorithms for simulation-based dynamic
traffic user equilibrium. Int J ITS Res 16(3):215–226. https://doi.org/10.1007/s13177-017-0150-6

123
Machine learning driven intelligent and self adaptive… 1217

19. Hamidi H, Kamankesh A (2018) An approach to intelligent traffic management system using a multi-
agent system. Int J ITS Res 16(2):112–124. https://doi.org/10.1007/s13177-017-0142-6
20. García-Nieto J, Alba E, Carolina Olivera A (2012) Swarm intelligence for traffic light scheduling: appli-
cation to real urban areas. Eng Appl Artif Intell 25(2):274–283. https://doi.org/10.1016/j.engappai.
2011.04.011
21. Jain P (2011) Automatic traffic signal controller for roads by exploiting fuzzy logic. In: Das VV,
Stephen J, Chaba Y (eds) Computer networks and information technologies, vol 142. Springer. Berlin,
Heidelberg, pp 273–277. https://doi.org/10.1007/978-3-642-19542-6_46
22. Hilmani A, Maizate A, Hassouni L (2020) Automated real-time intelligent traffic control system for
smart cities using wireless sensor networks. Wirel Commun Mob Comput 2020:1–28. https://doi.org/
10.1155/2020/8841893
23. Singh S, Vishwakarma S, Kispotta S, Yadav A (2020) A statistical analysis of various technologies
to detect and prevent fake news. In: Batra U, Roy NR, Panda B (eds) Data science and analytics, vol
1230. Springer, Singapore, pp 164–172. https://doi.org/10.1007/978-981-15-5830-6_15
24. Pop MD (2018) Traffic lights management using optimization tool. Procedia Soc Behav Sci 238:323–
330. https://doi.org/10.1016/j.sbspro.2018.04.008
25. Alexy AB (2021) Github repository for YOLO V3. https://github.com/AlexeyAB/darknet
26. PrashantJha SS, Kispotta S (2019) A lightweight approach for an adaptive traffic signal. Jabalpur, India
27. Liu X, Wang H, Dong C (2021) An improved method of nonmotorized traffic tracking and classification
to acquire traffic parameters at intersections. Int J ITS Res 19(2):312–323. https://doi.org/10.1007/
s13177-020-00247-w
28. Wang Z, Cui J, Zha H, Kagesawa M, Ono S, Ikeuchi K (2014) Foreground object detection by motion-
based grouping of object parts. Int J ITS Res 12(2):70–82. https://doi.org/10.1007/s13177-013-0074-
8

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

123
 Reproduced with permission of copyright owner.
Further reproduction prohibited without permission.