Distracted Driver Classification 🚗🚙🚌

1. Problem

Predict the Likelihood : What the Driver is doing in Each Picture?

We need to Classify Images into following 10 Classes

• c0: Safe Driving
• c1: Texting (Right)
• c2: Talking on the Phone (Right)
• c3: Texting (Left)
• c4: Talking on the Phone (Left)
• c5: Operating the Radio
• c6: Drinking
• c7: Reaching Behind
• c8: Makeup
• c9: Talking to Pessenger

2. Data

Kaggle's State Farm Distracted Driver Detection

3. Features

Data Set consist of Driver Images, each taken inside a Car with a Driver doing something.

• Texting | Talking on the Phone
• Operating Radio | Music Player
• Talking with Passenger
• Makeup
• Drinking
• Reaching Behind for some Object on the Back Seat.

4. Files

1. images.zip - All ( Train and Test ) Images.
2. sample_submission.csv - Sample Submission File in the Correct Format.
3. driver_images_list.csv - List of Training Images with Driver ID and Class ID.

5. Libraries

1. NumPy
2. Pandas
3. Matplotlib
4. Sklearn
5. Keras ( Utils )
6. Tensorflow ( Open Source Library to Develop and Train ML Models )

6. Approach

• Orignal Image Dimensions - 480 x 640 x 3
• Reduced Image Dimensions - 240 x 240 x 3
• Total Number of Images : 2242
• tensorflow.keras.utils.to_categorical : Converts a Class Vector ( Integers ) to Binary Class Matrix ( Dummy Encoding )

7. Parameters of Layers ( tensorflow.keras.layers )

A. Conv2D : 2D Convolution Layer

• filters : Number of Filters in Convolution.
• kernel_size : Size of Window.
• strides : How Far will Pooling Window Move | Steps.
• padding : valid ( no padding ) or same ( Add Same Padding from all sides )
• input_shape : Size of Batch or Window.
• activation : Name of Activation Function to be Applied ( e.g. activation = "relu" )

B. MaxPooling2D

• pool_size : Window Size ( Even if only one Integer is Passed it will be considered for both e.g. If 2 is Passed it is ( 2, 2 ) )
• strides : How Far will Pooling Window Move | Steps.
• padding : valid ( no padding ) or same ( Add Same Padding from all sides )

C. Flatten

• Pooling flattens the Input ( Down Sampling : e.g ( Input = ( 1, 10, 64 ) to Output = ( 640 ) i.e 1 x 10 x 64 = 640 )

D. Fully Connected Layer ( Feed Forward Connected Layer )

• Flattened Input is fed into the Fully Connected Layer.
• Takes Weighted Sum of all the Inputs from Previous Layer and Generates Output for Last Layer.

E. BatchNormalization ( Transformation : Mean = 0 and Standard Deviation close to 1 )

• Stabilize the Learning ( Reduces the Number of Training Epochs and Time required to Train the Neural Network )
• Network becomes Unbiased and makes Optimization Fast by Restricting Weights to Certain Range.
• Computes Means and Variance of the Feature in Mini Batch, then Subtracts Mean and Divides by its Mini Batch Standard Deviation.
• axis : Axis to be Normalized.
• epsilon : Variance Added to Prevent from Divide by 0 Error.

8. Performance Measure

• loss : categorical_crossentropy ( Probability Value between 0 and 1 ) and accuracy.
• Set Callback for Early Stoping to Prevent from Overfitting, monitor by val_accuracy and patience = 5 ( Loss <= 5 )
• batch_size : Training Data Set Size.
• epochs ( Number of Epochs ) : One Cycle of Full Training.

Accuracy : 97.51%