You are on page 1of 3

ABSTRACT

DEEP LEARNING BASED VEHICLE DETECTION AND COUNTER TO MITIGATE


HIGHWAY ROAD CONGESTION
By:
Izzan Dienurrahman

The purpose of this paper is to give a proposal of a system which intelligently detect vehicles on
a particular road and count them in order to give overview of traffic condition on a particular
highway, so that government can analyze and notify which highway route is better to go to.
Highway is an essential infrastructure to development of a country and industry, be it for
transporting things and/or person. However, in a developing country such as Indonesia
sometimes congestion occur. This phenomenon stalls transportation, therefore stalling the
development of the country and industry itself. On the other hand, government has built
alternative highway to solve this problem. Such problem still arises when people does not know
which highway is more congested than the other and chose the wrong road.
Artificial Intelligence/AI is a field where machine is made to think intelligently much like the
way human thinks and with that power, object detection can be done by the machine itself. As a
fast growing field and flexible uses of the technology, I believe that AI is the new trend to
approach all problems in the future.
The proposed system uses trained deep learning model to detects vehicle with webcam, and
OpenCV to draw a virtual line on a video frame to know whether the vehicle has crossed the line
or not, therefore the counting process can be done and best decision can be made after traffic
information condition is obtained.
Keywords: AI, deep learning, opencv, traffic congestion, vehicle detection

Background of the Problems


Highway congestion causes stalls in building the country developments, industrial shipment, and
overall people times. Such problem could be mitigated by choosing alternative highway route, in
order to do that, traffic information is needed to make that decision right. Sure there are other
ways to approach this problem such as using public transportation. In reality, public/mass
transportation has their own issues, such as cramped space, the security in public transportation,
the flexible mobility other than workplace, and not all workplace near the public station which
leads to more cost than using private vehicle. Based on those reasoning, people still prefer to use
private vehicle and congestion is the after effect due to popular choice amongst people.
Therefore, rising a challenge to overcome congestion problem.
Proposed Solution
The solution to this problem is to harness the power of AI, which enable machine to detect
objects if the machine is trained properly. First, we need to have deep learning model to detect an
object. In this case, we use SSD (Single Shot Multibox Detector) which as far as I know, has
high performance and accuracy. The next thing is to either train the model using a dataset
containing the vehicle images along with the class/label if it is a car/motorcycle or we can use
pre trained weight so we can just use the model right away to detect vehicles. Next, we use
opencv libraries to get live stream video input from a camera to feed image frame overtime to the
deep learning model.
The output of the model should contain objects that we want to detect based on the object
confidence level threshold. We can selectively filter to get a particular object and get the centroid
or center point of the object in coordinate (x, y) since we will draw each object with rectangular
box.

Centroid of a vehicle

Video input from: (Supercircuits, 2014)


Each centroid is saved inside a list, and when the next frame detection occurs causing a new
centroid on the image, that new centroid is compared to previous saved centroid. If the current
centroid position is below the virtual line (blue line) and the previous centroid is at the upper
line, therefore we count the vehicles by 1 and that newer centroids are saved, replacing previous
centroids.

At a glance, this seems easy and problem free, but in reality the detection may not stable causing
lost object in between frame detection. Lost object means lost centroid, and that previous
centroid needs to be compared to nothing because of lost detection. This causes invalid centroid
comparison to other vehicle centroid. This is where we use Pythagorean theorem, it calculates
the distance (Warehouse, n.d.) between previous centroid and newer centroid, and the least
distance will be matched with that particular centroid, creating indirect vehicle tracking, thus,
removing invalid centroid comparison.
With that information, government can analyze a lot of things, be it traffic suggestion, accident
investigation, highway development, etc.

References
Supercircuits. (2014, August 20). Alibi ALI-IPU3030RV IP Camera Highway Surveillance.

Warehouse, M. (n.d.). The Distance Formula. Retrieved from Math Warehouse:


http://www.mathwarehouse.com/algebra/distance_formula/index.php

You might also like