Skip to content

Youmin-Kim/2017-2MachineLearningProject

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 

Repository files navigation

2017-2MachineLearningProject

Using Convolution Neural Network to classify distracting driver's posture.


Abstract


Distracted driving habit is the biggest problem around the world. It causes many accidents and human disasters. To solve this problems, in this paper, I used the Kaggle dataset for “distracted driver” posture images with more distraction features than safe driving features. So I used machine learning method to identify driver’s posture while driving. Driver’s image has 10 classes which are drinking picture, texting posture, talking on the phone posture etc. My system used a supervised learning with general Convolutional Neural Networks (CNNs) and Softmax classification to classify this images into 10 classes. In this paper, this system shows 12% classification accuracy method.

DataSet


Kaggle, The data structure is as follows. The image of the total of 22424 driver’s driving is the 10 classifications. Image size is 640 * 480 pixels jpeg file. There are about approximately 2100 images according to each class. Each classes represent safe driving, texting with right-hand, talking on the phone with right-hand, texting with left-hand, talking on the phone with left-hand, operating the radio, drinking, reaching behind, hair and makeup, talking to passenger. Drivers are 100 persons.

Data Preprocess & Training


  • • I cut all 640 X 480 pixel 22424 images to 480 X 480 pixel images and resized to 192 X 192 pixel images. (Data Preprocessing)
  • • I made 5 training network layers consisted of 3X3 8 filters, convolution layer, relu function, 2X2 kernel size max pooling and drop out(keep probability = 80%). Each training network layer, number of filter increases twice. Initialized number is 8.
  • • I made 2 fully connected layers used classify images to their label. I used Adam Optimizer to minimize cost and softmax cross entropy to check costs. Accuracy function is comparing label’s argumax value and machine’s calculated arguma value.
  • • In training settings, learning rate is 0.01, data batch size is 100, 10 training epochs which means how many times to train all data.

Result


  • • I tested this model so many times but its result shows that its classification accuracy is 10% ~ 12%.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages