How well am I driving

The project ?How well am I driving? is based on the ideology of devising a system to check and rate the driving of a person  based on the data collected from the Can-bus via the sensors of the car and the visual data recorded with the help of a camera mounted on top of the car to gauge the

Project Title

How well am I driving

Project Area of Specialization

Artificial Intelligence

Project Summary

The project “How well am I driving” is based on the ideology of devising a system to check and rate the driving of a person  based on the data collected from the Can-bus via the sensors of the car and the visual data recorded with the help of a camera mounted on top of the car to gauge the  journey and the road specifications. This project will be specifically designed according to the traffic regulations and road conditions of Pakistan but can be later modified in order to serve different countries as well.

The project works on two separate domains which will then be later integrated to achieve optimization. The Can-bus data logger, deployed in the project provides readings for all the parameters such as speed, the temperature of the car, usage of indicators, usage of a car horn and some other sensing mechanisms. However, these readings will have to be decoded for further processing. In this regard, Honda Civic 2017 proves to be a good subject to experiment with since its decoder is available online and can be used to decode the datasets obtained from various sensors. Apart from the sensors, the front camera installed will give us valuable information on the behavior of the driver on the road. It will act as an aid to guide us how the driver maneuvers across different lanes and takes turns. Additionally, it will provide information on the distance between the subject car and the car traveling ahead and gauge the smoothness of the drive. Using these data combined will give us proper knowledge of how the driver was driving as the video will give us the data for the driving turning and the Can bus data logger will provide the information that if the driver used the indicators for turning while he was turning. Vision techniques will be used for object recognition motion detection and features will be extracted from both the datasets which will then be labeled. This labeled data will then be used to train a convolutional neural network which will then be used to assign ratings to the drivers in the real world. The weights associated with each feature will be adjusted until the accuracy can be further increased.

Project Objectives

The major goal of the project is to come up with SMART software that will be made available to the transportation industry in order to rank and assess their drivers. And in order to achieve this main goal the project has been broken down into some milestones.

Our aim is to target the transportation and freight companies and provide them with a tool to rank their employees and make it easy for them to grant bonuses and rewards based on the assessment done by our provided software.

Apart from the utility, the project aims to promote an accident-free safe environment by enabling the traffic police authority to gauge the deviations from an otherwise disciplined drive. The data set obtained after the implementation through Convolutional Neural Network will be then used as a standard set and a ranking system will be developed in this regard to compare and provide accurate results based on data received.

In a nutshell, the transportation and traffic authority in Pakistan is our main target audience which will benefit from the software developed under the course of this project.

Project Implementation Method

The project basically revolves around the idea of rating and improving the driving capabilities of a person. This will be achieved by collecting, visualizing and analyzing the data collected via a number of sensors mounted on the car and then decoding and optimizing it to get the best possible data set to be set as the training dataset. In order to achieve these following objectives have to be put into play:

1)         Data collection: This will be done by observing and collecting data from various subjects driving Honda Civic car with the pre-installed sensors sensing the parameters such as speed, the temperature of the car, usage of indicators, usage of a car horn and some other dynamic mechanisms. The other type of data collected will be the video from the GO PRO camera mounted on top of the car. GO PRO camera is chosen for this task specifically due to its water resistance, its video stability and its feature to be able to Geo-tag the pictures, which can be used in case of further studies.

2)         Analyzing the data: The data analyzation will take place after the required data is decoded using the CAN bus decoder available online. Once decoded the data will be classified in the following stage. The video data will be decoded using vision techniques of object recognition and motion detection.  A lot of research papers have been published by NVIDIA which talk about the algorithm used for object recognition and image detection and the computational power needed to run these algorithms in optimum time. These research papers will be very helpful in helping us implement these algorithms.

3)         Training of CNN: The decoded data will then be learned using CNN (Convolutional Neural Network). This aspect regarded as feature learning will primarily be used to evaluate the visual imagery obtained from the cameras mounted on the car in addition to the sensors.

4)           Classification: Once the feature learning is complete and a best fit available set of a standard is obtained the next and the last step is to classify and rate different driving behaviors based on this standard obtained through feature learning with will indirectly lead to a ranking of the driver under experimentation.

Benefits of the Project

This project will be converted into the software which will be commercially available to anyone who wishes to use it. It can be used by major transportation companies such as Uber, Careem in order to rate their drivers. The software will aid these companies in evaluating the drivers during the hiring process. Additionally, the software can also be used by heavy-weight freight companies or cargo companies to test the quality of their drivers and provide them rankings and bonuses based on that. It can also be used by the Traffic Police Authority in order to provide ratings for the fresh drivers who are applying for new licenses and give them the chance to improve and learn the driving techniques. However, in order to benefit from this model, the cars must be equipped with a front camera and the sensors.

The data collected during the span of this project will be clean and labeled which can afterward be used in further research projects.

The software in this project will be localized to working in Pakistan as it will be trained on the data that is specific to Pakistan’s traffic rules and driving conditions, but this software can, later on, be specialized for another company or another country just by gathering training data in those specific conditions.

Technical Details of Final Deliverable

The final deliverable, in this case, will be a software that is well integrated with a GO PRO camera and the CAN-bus reader, which relates to the sensors of the car. The software will be able to decode data from the CAN-bus reader and will also be able to extract features automatically from the camera output. It would then make predictions about the rating of the driver throughout the course of his drive using the trained CNN architecture and the weights associated with it.

Since the feature extraction requires objection recognition and motion detection and video processing, therefore it would require high computational power which will require some hardware installation to carry out these processes. The software will be run on a GPU which will be a part of small custom-built hardware specially designed for the project. The hardware will be connected to the camera and the CAN-bus through connecting wires for smooth and continuous transfer of data. This piece of hardware will get the power source from the car battery and some experimental modifications will be made in order to ensure that the device doesn’t get heated up and performs its functions without any drawbacks.

Final Deliverable of the Project

Hardware System

Type of Industry

Transportation

Technologies

Artificial Intelligence(AI)

Sustainable Development Goals

Reduced Inequality

Required Resources

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