الفهرس 
Detection-Based MethodsOnly 14 pages are availabe for public view
 |  | from | 122 |  |  |
| | | from | 122 | | |
Abstract
ii
Abstract
In highway management, intelligent vehicle detection and counting are
becoming increasingly important as an accurate estimation of traffic
density on road congestion reduction. Traffic density estimation is
affected by the difficulties of perspective distortion, size change,
significant occlusion, and background interference in traffic images. To
address the previous issues, a model that enhances the quality of
estimating traffic density is developed. The efficientNet fine-tuning
architecture is used then, followed by the development of seven dilated
convolutional layers to extract the deeper features in the images that
maintain the output‟s resolution to generate a high-quality density map.
Finally, the vehicle count will be calculated from the high-quality density
map. The experimental results indicate that the suggested approach
significantly enhances the accuracy of traffic density estimation compared
to the existing ones. It achieves promising results on the TRANCOS
dataset with a Mean Absolute Error (MAE = 5.23) and Grid Average
Mean Absolute Error (GAME (1)) = 5.39, GAME (2) = 5.52, and GAME
(3) = 6.40).This is the first phase in this thesis that will be used as input to
the second phase, which aims to predict short-term traffic flow to solve
the problem of congestion on roadways; that is an issue in many cities,
especially at peak times, which causes air and noise pollution and cause
pressure on citizens. So, the implementation of Intelligent Transportation
Systems (ITSs) is a very important part of smart cities; as a result, the
importance of making accurate short-term predictions of traffic flow has
significantly increased in recent years. However, the current methods for
iii
predicting short-term traffic flow are incapable of effectively capturing
the complex non-linearity of traffic flow that affects the prediction
accuracy. To overcome these problems, a hybrid Long Short-Term
Memory (LSTM) and hybrid Gated Recurrent Unit (GRU) neural
networks were presented on the basis of the LSTM model and the GRU
model. Hybrid LSTM and hybrid GRU models were evaluated in
comparison to the other usual models. It is discovered that the hybrid
LSTM model and the hybrid GRU model have demonstrably less
prediction error than the other models. The experimental results show that
the Mean Absolute Error (MAE=7.14 for hybrid LSTM and MAE=6.98
for hybrid GRU), the Mean Absolute Percentage Error (MAPE=16.73 for
hybrid LSTM and MAPE=16.50 for hybrid GRU) and the Root Mean
Square Error (RMSE=9.74 for hybrid LSTM and RMSE=9.51 for hybrid
GRU) on PEMS dataset and introduce promising results on TRANCOS
datasets. Hence, a hybrid LSTM and a hybrid GRU can extract the traffic
flow temporal features and are able to suddenly capture trend chang