Training Self Driving Cars by leveraging Raspberry pi sensors and neural networks

Autonomous cars have now gone uphill from science fiction to reality. Basic technologies behind the self-driving cars are deep learning and computer vision techniques. Self-driving cars can potentially overcome the mistakes made by human drivers thus saving human. In this project, a pr

Project Title

Training Self Driving Cars by leveraging Raspberry pi sensors and neural networks

Project Area of Specialization

Robotics

Project Summary

Autonomous cars have now gone uphill from science fiction to reality. Basic technologies behind the self-driving cars are deep learning and computer vision techniques.

Self-driving cars can potentially overcome the mistakes made by human drivers thus saving human. In this project,

a prototype of self-driving is developed that has the ability to maneuver on predefined paths using convolutional neural networks and computer vision techniques.
 

Project Objectives

• Development of working prototype of a self-driving using a toy car that will mainly navigate using computer vision and deep learning techniques
• Usage of Convolutional Neural Network to identify a stop sign
• Collision avoidance of self driving car
• Object detection infont of self driving car
• Recognition of speed limits on speed-limit signs by using Convolutional Neural Network
• Car sound horn and car brake lights

Project Implementation Method

First Collect the data then pre-processing, the data is feed to a CNN model for training. 
The training and inference is done using core i5 laptop and 1070 core i7 system. 
Training data used in our project is about 70 percent of the complete data.
 Supervised learning is used for training of the data.
 This data is classified and labeled as Right, left and forward. This data is trained using CNN sequential model.
 CNN model used for the training of the data contains 15 hidden layers. These layers include dense layer, convolutional-2D layer,
 maxpooling-2D layer, flatten layer and fully connected layers. CNN is used for extracting the features from the images and learn 
through these features by updating the bias and weights of the perceptron. Categorical cross entropy with Adam optimizer and a learning 
rate of 0.001 is used in this model. The trained model then takes the input images from live camera and predicts which direction to choose
 or stop. The trained model after prediction generates a string and through serial communication the string is sent to Arduino. Finally, 
the Arduino processes the wrappers embedded in its code according to the string received from the trained model and sends control signals
 to the H-bridge to drive motors of the car to move or stop according to the prediction.

Benefits of the Project

• Prevention of car crashes
• Societal cost-savings
• Traffic efficiency
• Better access and mode of transportation
• Environmentally friendly
• Reduction in traffic deaths
• Drop in harmful emissions 
• Lower fuel consumption

Technical Details of Final Deliverable

Software
Python is used in this Project as programming language

Libraries are requried for this project

• OpenCV
• Numpy
• Matplotlib
• scikit-learn
• TensorFlow
• Keras
• Pickle

Hardware 

• Toy RC Car 
• Rechargeable Batteries
• H-bridge Motor driver
•  Arduino Mega 2560
•  Pi Camera Webcam
• Training tracks
• Raspberry Pi
• Powerbank

Final Deliverable of the Project

Hardware System

Core Industry

IT

Other Industries

Core Technology

Robotics

Other Technologies

Artificial Intelligence(AI), Internet of Things (IoT)

Sustainable Development Goals

Good Health and Well-Being for People

Required Resources

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