Research

On the Internet of Connected Vehicles, a vehicle has to communicate bi-directionally with several devices for establishing a shared network for inter-vehicle and intra-vehicle connectivity. These connection protocols are commonly structured to connect all the individual components with an implicit degree of trust, which is supposed to protect the whole system from unauthorized users. Technologies like Automotive Ethernet tend to increase security by reducing the implicit trust within the local network devices. However, the lack of individual security protocols in vehicle-to-vehicle communication still keeps the possession of vulnerability to hacks, external attacks, and further disruption.

This is where Zero Trust Architecture can become a reliable technology for the exchange of information between vehicles. Zero trust is a security system that means no one is trusted by default and verification is required from anyone or any device willing to get connected to the intra-vehicle network.

(--- Further details will be published later ---)

This report is an analysis of a particular dataset that is generated to estimate the goal of an ultra-low-power cyber-physical system called the PhyNode. It is an embedded computing and sensing platform (wireless sensor network node) in the PhyNetLab which is deployed in real-time materials handling system. The function of PhyNode is to deliver 3 values from the sensors. These values can be used to understand the specific position of the PhyNode and to navigate the material handling operator towards the target. The values are:

1) Light intensity

2) Acceleration in x, y, z-direction (Towards floor, corridor, human operator respectively)

3) Temperature

My report focuses on analyzing the particular dataset that is generated to estimate the goal PhyNode. The major part of the estimation involves working on developing a machine-learning algorithm to analyze the dataset and learning one specific position of a PhyNode.

I studied the dataset and identified it as a classification problem. To solve the task, I used the Random Forest classifier and eXtreme Gradient Boosting classifier, two effective machine learning classifiers for this specific dataset. The methodology I followed include data pre-processing, data visualization. The models gave 98.05% and 97.88% accuracy respectively. I also built a feedforward artificial neural network (ANN) to establish a deep learning classifier for the dataset. This model provided an amazing score of 99.94% test accuracy overall.

Traffic sign detection and recognition are modern -aged technological development of intelligent vehicles. Even though there is no implementation of this technology in our country, the need for autonomous traffic navigation in the streets is needless to say. The early stage of traffic-sign detection and recognition algorithms face some problems such as traffic signs not being available on the roads, change in the state of traffic lights, a low-resolution camera on the vehicles, road-side signs not easily detected, and poor real-time performance of deep learning-based methodologies for traffic-sign recognition. In this work, we proposed an improved traffic-sign detection and recognition algorithm named YOLOv3 using OpenCV in python language.

We aimed to address the problems such as how easily affected traditional traffic sign detection is by the environment. Our work with the YOLOv3 algorithm showed real-time performance for traffic sign recognition and addressed how it can help to reduce the number of road accidents in Bangladesh. The overall methodology exhibits real-time object detection and classification at around 45 frames per second with 92.2% accuracy. We can use this algorithm to overcome all kinds of sign detection and recognition problems we have faced and use them in our country to improve intelligent vehicle systems.

Bangla handwritten character recognition is a difficult job compared to other languages due to the morphological complexity of adjacent characters and a wide variety of curvatures in writing styles people have. Another reason for that is the unique presence of compound characters. Most of the recent research works conducted in this field standardize Deep Convolutional Neural Network (DCNN) models for delivering the most effective outcomes. This paper proposes a DCNN model to classify all the character classes from three popular databases known as BanglaLekha Isolated, Ekush, and CMATERdb. As for BanglaLekha Isolated, our model achieves 93.446% accuracy on the 50 alphabets category and an overall 91.45% considering the whole dataset. The other two datasets, Ekush and CMATERdb result in 95.05% and 94.17% respectively, where the second one holds 171 classes of compound characters alone and performs 93.259% correctness, which is so far the best for this specific category in this dataset.

The research aimed to improve the basic structure of OCR-based identification of documents and exam papers for the educational institutes with Bangla medium. The future modification of this research is to focus on detecting sequence of characters or words or sentences in Bangla Handwritten documents and implement it as OCR based identification system.