Project 5 / Face Detection with a Sliding Window
Training Data
The first part of the detection requires us to create a HOG feature. We achieve this by creating a list of features with labels from both positive(faces) and negative(no faces) images. Then we use a linear SVM to train the features and runs it against a test. SVM works well due to the fact the data is high-dimensional.
Sliding Window with Non-max Supression
To detect faces in images, we used a sliding window with a combination of various sizes of frames. We get the HOG features for each frame and fit it with the linear SVM modal. If the confidence is above the threshold we include the frame as a positive frame (is a face).
Results
Face template HoG visualization
