e7dud7e / carnd

Self Driving Car nanodegree - Udacity

Self-Driving Car Engineer Nanodegree

Overview of projects

This is an overview of self driving car term 1 projects that I completed for the Udacity Self-Driving Car Engineer Nanodegree.

Term 1 covers computer vision using traditional methods as well as with neural networks.

Install

The projects use either OpenCV or TensorFlow. See the individual README of each project for setup details.

Lane Detection

See the project writeup ./LaneDetection/LaneDetectionWriteup.md

See also the videos the lane detection here: LaneDetection/test_videos_output/

Use OpenCV to detect lane lines in video from a mounted camera of a driving car.

Traffic Sign Classification with CNNs

• See the project writeup ./TrafficSigns/project_writeup.md.
• Built convolutional neural networks to classify traffic signs, and reach 97.55 test accuracy.
• Used image augmentation and batch normalization to improve the model.

Behavioral Cloning with CNNs

• See the project writeup in writeup_report.md
• See the video.mp4 for a test drive based on the trained network (in a simulator).
• Used Keras to build a convolutional neural network that predicts steering angles using video images.

Advanced Lane Detection

• See ./LaneAdvanced/advanced_lane_writeup.md

• See the video ./LaneAdvanced/output_video/project_video_out.mp4

• Used OpenCV to undistort camera images, used perspective warp to generate an aerial view of the road.

• Used gradients and color spaces to detect lane lines and also detect the entire lane within lane lanes.

Vehicle Detection

• See the project writeup in ./VechicleDetection/project_writeup.md
• See the videos in ./VehicleDetection/output_video
• Used histogram of gradients to extract features from images.
• Used a rolling window and SVM to classify areas of the image as a car or not.
• Tracked detected images over time in video.