This project implements the method of 'Bag of Visual Words' to classify images into vehicles and non-vehicles from the available dataset.
Following are the steps implemented (in order):
- Split the dataset into Training and Test (already divided the dataset in the respective folders)
- Extract SIFT features from the training dataset
- Learn a codebook of size 1000 i.e. the size of Bag of Visual Words is 1000, using K-Means clustering
- Train a linear, soft-margin SVM using 5-fold cross-validation on the training dataset
- Test the accuracy of the SVM model learnt on the test dataset
Below is the plot of accuracy (during cross-validation) against the value of regularisation parameter:
Classification Accuracy on Test Data: 83.65%