Deep Learning based Visual Computing Solution for Monitoring Vehicular Traffic

Deep Learning based Visual Computing Solution for Monitoring Vehicular Traffic in Mehran University Includes the following things: Monitoring access of vehicles into MUET. Tracking vehicle movement and traffic monitoring. Restricting unauthorized access

Project Title

Deep Learning based Visual Computing Solution for Monitoring Vehicular Traffic

Project Area of Specialization

Artificial Intelligence

Project Summary

Deep Learning based Visual Computing Solution for Monitoring Vehicular Traffic in Mehran University Includes the following things:

• Monitoring access of vehicles into MUET.
• Tracking vehicle movement and traffic monitoring.
• Restricting unauthorized access

The increasing need for security has called for improved methods for restricting access to un authorized vehicles into the university. It further tracks the movement of vehicles in and out of the campus that could be tracked online anytime. It will further monitor traffic density and provide statistics for proper planning and prediction for increasing or decreasing vehicles playing at any particular time of the day.

Vehicular traffic will be tracked as it moved initially into MUET to monitor the traffic by identifying its movements.

The unauthorized and non-recognized vehicular traffic will be restricted and its being recognized by the number plate of vehicle.

 As a Case Study we are implemention it into Mehran University, but the larger scale implementation will be in Hyderabad City. 

Project Objectives

Monitoring movement of any organizations fleet is important for smooth operation. It not only improves the efficiency of the system but also controls unnecessary fuel consumption especially when the organization is located at a faraway location and not accessible to all stake holders on their own conveyance. Mehran university has a large fleet of buses plying between university and different cities. Therefore, we aim to develop a visual computing framework for monitoring various aspects of vehicular traffic in Mehran University’s Jamshoro campus which provide effectual traffic control conditions and solve problems such as traffic congestion and traffic accidents. To develop such embedded system solution for vehicular system which yields intelligent transportation system.

The main project objectives are stated below.

• To Understand Deep Neural Network architectures for Visual Computing.
• To Implement Deep Neural Network models on Raspberry Pi 4.
• To interface camera with Raspberry Pi 4 and processing of test videos in real-time.
• To develop a Real-time survelliance system for monitoring vehicular traffic entering and exiting thhe university gate.
• Multiple-object Sensing: To accurately provide the traffic flow. This embedded system need to be capable of sensing multiple objects.

Object Classification: In the real world scenario, there will be many types of objects picked up by the camera. Consequently, the intelligent transportation embedded system needs to be able to extract vehicles from other types of moving objects.( objectives which are written in itatic form are optional.)

Project Implementation Method

Implementation will be based on an embedded system which will be designed in a way to follow the following methodology:

1. At the very first “Raspberry pi Cam” will read the video of vehicular traffic.
2. On the second phase “Deep Learning Vehicle Detection” methods will be implemented on that video.
3. Vehicular traffic counting as it moves through the particular predefined area by using few Deep Learning algorithms (YOLO: You only look once)
4. With help of video frames, the number-plate recognition will be done, by using few defined algorithms of deep learning ANPR (Automatic number plate recognition) and ALPR (Automatic License plate recognition) of vehicular traffic.
5. The speed estimation will be done, by using Deep Learning Algorithms as the vehicle changes its position per frame.
6. All above algorithm’s results will be collected and forwarded to the user interface.
7. At the very last stage of Tracking the Vehicle each vehicle will be assigned an ID with the help of which the vehicle will be easily tracked into the network by following the deep learning Tracking algorithms.

Benefits of the Project

Numerous benefits of project are for vehicular traffic and as well as for the pedestrians. Following benefits are listed sequentially

1. Tracing vehicles. The system will help the university administration to monitor round trip time of any bus and come up with statistics for better control on fuel consumption and efficiency.
2. Security. The system will allow or deny the authorized or unauthorized vehicle into MUET campus resulting in a safe and secure environment. A check on unauthorized vehicles not only prevents any law and order situation but also restricts any threat to others especially pedestrians due to possible violation of traffic rules.

Speed estimation. The system will monitor the speed limits on campus and puts check on speeding vehicle for safety.

Technical Details of Final Deliverable

Having the best efficiency of Deep Neural Network for detection using the different algorithms, the raspberry pi 4 4Gb kit along with the 5MP raspberry pi Cam can work the best way to detect, monitoring and tracking the vehicular traffic into Mehran university.

The algorithm YOLO (You Look Only Once) object detector consists of CNN called Darknet, YOLO v3 the prediction is performed at different scales. YOLO algorithm is fast because it has requirement of only one image and splits it into an NxN grid where each cell predicts a fixed number of bounding boxes to associate an object to the supposed class providing a confidence score [1]. Traditional machine vision methods use the motion of a vehicle to separate it from a fixed background image. This method can be divided into three categories i) the method of using background subtraction ii), the method of using continuous video frame difference iii), and the method of using optical flow iv). Using the video frame difference method, the variance is calculated according to the pixel values of two or three consecutive video frames. [2]

Object Counting Every passing vehicle object inside ROI (Region of Interest) will be tracked based on its position and would be compared with the list of tracked object positions. For a new position or position not including in the list of tracked objects, it will be added as a new object and should be counted. If the new position was included in the list of positions of previous tracked objects, it means the position had already been counted as a recognized vehicle.[3]

References:

[1] Impedovo, Balducci , Dentamaro, Pirlo “Vehicular Traffic Congestion Classification by Visual Features and Deep Learning Approaches: A Comparison” pp.3-5 28 November 2019.

[2] Song, Liang, Huaiyu, Dai & Yun “Vision-based vehicle detection and counting system using deep learning in highway scenes” European Transport Research Review volume pp.3-4 30 December 2019 .

[3] Ramadhani, Eko Minarno, Budi Cahyono “Vehicle Classification using Haar Cascade Classifier Method in Traffic Surveillance System” Universitas Muhammadiyah Malang KINETIK, Pp. 57-66, Vol. 3, No. 1, February 2018

Final Deliverable of the Project

HW/SW integrated system

Core Industry

Transportation

Other Industries

Others

Core Technology

Artificial Intelligence(AI)

Other Technologies

Artificial Intelligence(AI)

Sustainable Development Goals

Quality Education, Industry, Innovation and Infrastructure, Life on Land

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com