/Driver-Drowsiness-Detection

Primary LanguageMATLAB

Driver-Drowsiness-Detection

List of scripts:

Main Pipeline Example

Preprocess, reduce and train classifier for one PML technique. It makes use of the following functions:

  1. PreprocessDataset: collects all videos from each training/test dataset subject and situation and applies the desired PML preprocessing.

  2. BuildTrainDataset: Builds a dataset from a file created by PreprocessDataset

  3. DatasetPCA: Performs PCA analysis on given dataset

  4. ReduceDataset: Reduces a dataset using the structure returned by DatasetPCA

  5. TrainReducedDataset: loads the reduced training and test datasets and build SVM classifiers modifying the number of training features, the SVM kernel parameters and the SVM C parameter. A file with relevance data from the result of each trained classifier is saved as Results_ReducedPCA'PCA_Ratio''kernel'_Sub'subsampling'.mat. This file contains a data structure with the following fields:

    • situation
    • kernel
    • kernelScale: kernel Scale applied to the SVM classifiers
    • FisherPerc: percentage of features according to Fisher Score
    • TestAccuracy: accuracy of classifiers on test dataset for the combination of the two previous parameters
    • TrainAccuracy: accuracy of classifiers on train dataset for the combination of the two previous parameters
    • confusion_matrix: confusion matrix for each classifier
    • max_coordinates: best percentage of features to be selected and best kernelScale.
    • BoxConstraint: values of C applied to the SVM classifiers once the number of selected features number is fixed based on results on previous field.
    • tunedKernelScale: kernelScale applied to the SVM classifiers once the selected features number is fixed.
    • tunedFisherPerc: best percentage of features selected
    • tunedTestAccuracy: accuracy of classifiers on test dataset for the combination of BoxConstraint, tunedKernelScale and tunedFisherPerc.