Abstract:
In urban cities, tra c management of intersections is a substantially challenging prob-
lem. In appropriate tra c control leads to waste of fuel, time, and productivity of
nations. Though the tra c signals are used to control tra c, it often causes problems
due to the pre-programmed timing being not appropriate for the actual tra c intensity
at the intersection. Tra c intensity determination based on statistical methods only
gives the average intensities expected at any given time. However, to control tra c
e ectively, the knowledge of real-time tra c intensity is a must-have. In this project,
vision-based technology and arti cial intelligence (AI) are used to estimate tra c in
real-time and control the tra c in order to reduce the tra c congestion. General
-purpose electronic hardware has been used for in-situ image processing based on edge-
detection methods. A Neural Network (NN) was trained to infer tra c intensity in each
image in real-time using a scale of 1(very low) to 5 (very high). A Trained AI unit,
which takes approximately 4 seconds to process each image and estimate tra c inten-
sity was tested on the road where it recorded a 90% acceptance rate. In order to control
the tra c, a ratio-based method and a reinforcement learning (RL)-based method was
used. The performance of these methods are compared with a pre-programmed tra c
controller.
