Fine Grained Recognition of Road Traffic Vehicles

The advancement in technologies and the rapid growth rate in urban population have led to problems in the urban transportation network. In this day and age the metropolitan areas of the world consist of more vehicles than ever. In case of Pakistan the car financing schemes and more economical option

Project Title

Fine Grained Recognition of Road Traffic Vehicles

Project Area of Specialization

Computer Science

Project Summary

The advancement in technologies and the rapid growth rate in urban population have led to problems in the urban transportation network. In this day and age the metropolitan areas of the world consist of more vehicles than ever. In case of Pakistan the car financing schemes and more economical options have helped buy numerous people buying their own vehicles. As a result we have seen a rise towards traffic congestion, more accidents and loss of human lives. One of the leading cause of accidents in Pakistan is reported not a disease or an act of terrorism but loss of lives in car accidents. These accidents are a result of many factors including unsatisfactory road infrastructure. A solution to these problems lies in a better transportation system. More and more countries have adopted intelligent transportation systems as a result. An intelligent transportation system (ITS) is an advanced application which aims to provide innovative services relating to different modes of transport and traffic management and enable users to be better informed and make safer, more coordinated, and 'smarter' use of transport networks. This is all done using Computer Vision. The intelligent functions of current intelligent traffic video monitoring systems focus on object detection and tracking. This paper proposes an approach based on convolutional neural networks (CNNs) combined with transferal learning and fine tuning. In order to identify and classify the vehicles locally present. To achieve this first one of a kind local data set is also constructed consisting of local types of vehicles. As a result, more work can be performed in the domain using a unique dataset.

Project Objectives

The aim of our research is to propose an algorithm responsible for granular level of analysis on the local dataset. It will be responsible in recognizing different vehicles and there models. For example, Corrolla, civic, Aqua, Cultus and different design vehicles. In addition we aim to differentiate between the yearly models aswell such as 2007 corolla and 2020 corolla. This would only be possible by preparing data set that consist of such vehicles from qinchi to land Cruisers.

Project Implementation Method

The first step is to achieving the required results is to establish a standardized publishable data set. This data set should be in accordance with international standards so researchers from all over the world can use it for more ground breaking researches. Due to its uniqueness it’s not to be found yet on any dataset collection. The data set would contain the labelled images of the types of vehicles found in the local environment i.e. Karachi, Pakistan. The types consist of majorly sedans, hatchbacks, public transport busses, motor bikes, chinqi and rickshaw. It will be a mini data set but enough to be used for mini data set machine learning models.

After establishing labelling the 70 percent images of the data set will keep the 30 percent for further testing the algorithm towards the end.

Using Convolutional Network machine learning models we will train the first few layers of the CNN so the mode can differentiate between the basic car types i.e. sedans, hatchbacks, coupes. The last two layers will used for transferred learning and lastly fine tuning of the model. After training of the first few layers we will freeze those layers and not change them in the future, the new data set will train the model further for our required results after thorough training the fine tweaks of the model will be done through which we hope to achieve decent accuracy level.

Benefits of the Project

There are many benefits of the project. First and foremost is its application in computer vision system ie intelligent transport System such as the one that central development authority , Islamabad has initiated on the expressways. The data set of the local transportation of the Country and the types of cars will easily be relatable to the third world and developing countries. 

Another benefit is its relevence for data collection for roady infrastructure development companies.Companies need the detat about articular and the type / flow of traffic. this in turn decides the road thickness , bus stops, lanes, speed breakers etc. A real time security detection system can be further developed based on vehicle tracking. Numerous application of convolutional networks based approaches to solve the daily transportation problem are all possible.

Technical Details of Final Deliverable

The result of the identified cars and the number found will be displayed on a dashboard. After training we can use our Model on testing data to recognize the vehicles present. Thus the model could then recognize vehicle classifying them in to sedans, hatchbacks, heavy trucks. Also the models civic, corolla, aqua, cultus etc.

Thus a dashboard we video feed willl be able to recognise vehicels at real time a well as from any samle.

A data set that can be usef for further researches. Consisting of all local transport vehicles

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

Industry, Innovation and Infrastructure, Sustainable Cities and Communities

Required Resources

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