This thesis entitled "Intelligent Traffic Management System and Congestion Analysis Using Artificial Intelligence" submitted by Nishat Ahmed Samrin, ID Department of Computer Science and Information Systems, Daffodil International University, has been accepted as satisfactory in partial fulfillment of the degree requirements B.Sc. It was my pleasure to research and complete my thesis on "Intelligent Traffic Management System and Density Analysis with Artificial Intelligence". I am grateful to Allah for His grace in my education. I would like to express my gratitude to my supervisor, Abdullah Bin Kasem Bhuiyan, Lecturer in CIS Department, Daffodil International University, for his valuable time, advice and inspiration to give his best to this paper and his education.
I would like to thank my sir and send my best wishes to him for his faith, steadfast guardianship, sound judgment, suitable outlook and understanding of various factors that helped make my thesis possible. He did everything in his power to ensure that I knew what was needed to complete this thesis. Sarwar Hossain Mollah, the head of my department, for his helpful help at the beginning of the study and making sure that I understood the subject.
Mehedi Hassan, lecturer in CIS department at Daffodil International University, and all my lecturers for their unfailing help and patience in getting this thesis work done. For improved signal management and efficient traffic control, it is also important to determine real-time traffic density on the roads due to its ever-increasing nature.
1
- Introduction
- Research Objective
- Motivation
- Rationale of the Study
- Research Questions
- Expected Output
As a result, predicting traffic density levels based on these variables enables decision makers to make informed decisions and take appropriate actions to address problems. A model performance analysis is done to understand the predictive power of the proposed model. In the proposed system, I have used the vehicle detection module, Signal Switching Algorithm, Yolo5, Vehicle Calculation and Signal Timing Variation to predict the traffic density and give the best solution for traffic management.
Various parts of this paper are organized below: Problem background, research objectives, motivation, research questions, expected output result, literature reviews of the existing works, the scope of the problems, challenges of this implementation, Proposed system overview, vehicle detection algorithm, Signal detection item, implementation requirement, operation of the algorithm, simulation module, use of Yolo, Experimental results and discussion, comparison of models, performance evaluation, Conclusion and Future Works in the last section. To develop the best possible intelligent traffic density control and monitoring system using computer vision. I also saw that many researchers were already working on their theoretical traffic solution research, but there was very little research on traffic density.
Traffic research is very difficult due to proper tracking, data and many other things. Throughout this research, we will analyze traffic density and develop an automated system for better traffic management.
5
- Introduction
- Related Work
- Bangladesh Perspective
- Scope of the Problem
- Challenges of this Work
On the other hand, (Osman et al., 2017) we refer to; they aim to make an advanced driving traffic system with computer vision for a cross section of roads. Their proposed system would count the number of vehicles and determine the ideal time to wait before flashing a red signal and running a green signal based on the number of vehicles. As technology advances, they can now collect data from the Google Maps traffic layer for the least amount of money.
This article first collects data from Google Maps traffic layer and then automatically captures custom traffic layer images. To estimate the density, they update the times, focusing on the composition of the dataset, the size of the moving horizon and the congestion criteria. (Nemade, 2016) The proposed approach addresses key traffic issues while providing tailor-made solutions for Indian roads and automobiles. Then we choose to manually collect and supplement data from the internet, so that there is sufficient data volume to train a YOLO model.
Since it takes many hours to train a model on the local CPU, we looked for a solution since we didn't have a high-end GPU. Eventually we discovered a solution called Pygame, Google Collab and Visual Studio, and used a paid cloud GPU to quickly train our model.
8
- Proposed Traffic System Overview
- Data Collection of Traffic
- Data Preprocessing
- Data Augmentation of Traffic
- Data Labelling of Traffic
- Vehicle Detection Module
- Signal Switching Algorithm
- Implementation Requirement
- Simulation Module
- Real-Time Module
The remaining 20% of the data was used for testing and the remaining 80% for training. To reduce the number of objects in the graphic, we first cropped the entire image in a variable four-dimensional dimension. By improving the image classification process, significant overfitting and underfitting of the model to the training data can be avoided.
By manually labeling images scraped from Google and added to the dataset of traffic for the model's training, LabelIMG. Next, pre-trained weights obtained from the YOLO website were used to train the model. The setting of the .cfg file used for training has been changed according to the specifications of our model.
By changing the 'classes' variable, the number of output neurons in the last layer was changed to match the number of classes the model should recognize. After loading the model and inputting an image, the results are output in a JSON format or as key-value pairs, where labels serve as keys and the values are the trust and coordinates of the labels. The algorithm divides the image into regions, predicts the boundaries and probabilities for each region, and applies a single neural network to the entire image.
By changing the 'classes' variable, the number of output neurons in the final layer was changed to match the number of classes the model is intended to recognize. After loading the model and feeding an image, it outputs the results in a JSON format or as key-value pairs, where labels serve as the keys and the values are the confidence and coordinates of the labels. The key is the label of the detected item, and the data in JSON format is the confidence and coordinates.
Any number of signals at an intersection can be accommodated by scaling the algorithm up or down. NumOfVehicles = The number of cars, bikes, CNG, trucks and buses in each class NumOfVehicles represents the vehicle detection module that counted at the signal. Also displayed next to each light is a number of vehicles that have passed through the intersection. Vehicles of various shapes and sizes are cars, buses, bikes, trucks, CNGs and others arrive from all angles.
Some vehicles in the far right lane turn and cross the intersection to increase the realism of the simulation. Here are all times, default red 150, default yellow 5, default green 20, default minimum 15, number of signals is 4.
20
- Introduction
- Experimental Dataset
- Evaluation of Proposed Traffic System
- Performance Evaluation
- Result discussion of model
- Comparative models of object detection
- Comparison to some previously completed works
- Comparison with static system and proposed system
In our dataset, 80% of the images were training sets and the remaining 20% were test sets. We coded and trained our model entirely on a Google Collab notebook using the premium cloud GPU. On the training set, our proposed system provides an accuracy of 86 percent, and on the test set an accuracy of 85 percent.
Recall: Recall heps to measure the proportion of positive values between correctly classified and all actual class yes observations. We know that this system uses several techniques, including trajectory-based, GPS, FASTER R CNN, Yolo-V3, and R CNN.
34
35
Prediction of traffic density using YOLO Object Detection and implemented in Raspberry Pi 3b + and Intel NCS 2. Implementation of an Unreal Engine 4-Based Smart Traffic Control System for Smart City Applications”, International Journal of. Detection and implemented in Raspberry Pi 3b + and Intel NCS's International Conference on Vocational Education and Training (ICOVET), 2020.
