Modular Package for Autonomous Driving (mPAD)
As part of my senior project at Worcester Polytechnic Institute, I led the development of a Modular Package for Autonomous Driving (mPAD), designed to enhance autonomous vehicle capabilities.
Key Achievements
Advanced Pre-processing Methods
Implemented a sophisticated pre-processing method that reduced noise by 90% using a combination of techniques:
- Grayscale transformation
- Gaussian blur
- Canny filter
- Hough line transform
Additionally, I introduced artificial noise to facilitate easier pattern identification, improving the system's ability to recognize road features.
Object Detection with Parallel Processing
Incorporated object detection with parallel processing by employing a Convolutional Neural Network (CNN) model. This implementation achieved an impressive frame rate of 20 frames per second, enabling real-time detection capabilities essential for autonomous driving.
Integration with DonkeyCar
Successfully integrated the pre-processing and object detection components into the production code for the open-source platform DonkeyCar, enhancing its capabilities and performance.
Reliable Autonomous Operation
Achieved reliable autonomous driving after just 5 laps of manual driving, with robust object detection and comprehensive sensor data collection, demonstrating the effectiveness of the implemented solutions.
Technologies Used
- Computer Vision
- Convolutional Neural Networks (CNN)
- Parallel Processing
- DonkeyCar Platform
- Image Processing Techniques
- Sensor Data Integration