Goal:

• Estimate a bounding box for detected vehicle .

Process:

1. Color Feature
2. Histogram of oriented gradients(HOG) Feature
3. Build Vehicle Classifier
4. Window Search

The final output comes out to be:

1] As cars tend to have stronger Saturation values from HSV color space than non-cars, I have used its color values as part of feature.

2] The idea behind HOG feature is to identify shape of gradients by blocks in a given channel of image. Through HOG extraction, we are interested in distinguishing a car image from a non-car by looking at its edges. 'get_hog_features' takes input of image, number of orientations, and size of cell and block. HOG will aggregate cell values and determine the gradient for each block.

After extracting features from all data, I used SVM to train the features. Before training the data, the data was normalized using StandardScaler() from sklearn.preprocessing. Then these normalized data were splitted into train and test sets. (80% Training, 20% Test set).

I trained a linear SVM using LinearSVC from sklearn and got test Accuracy of SVC = 0.987.

• Implement a sliding-window technique and use the trained classifier to search for vehicles in images.

With trained vehicle classifer, I would run classifier across pixels of image. First, I would get multiple detections for a car (image1). Then, I create a heatmap to count and draw a box with endpoints that fit those detections. (image 2, 3) Finally, I would draw the box on original image to output a clean outlook of vehicle detection. This method prevents false positives and multiple detections to be shown in image.