project05.frame module

class project05.frame.Frame(img=None, imgpath=None, min_row=0, max_row=None, scale=1.0, spatial_cspace='BGR', hist_cspace='BGR', hog_cspace='BGR', spatial_size=32, hist_bins=32, hist_range=(0, 256), hog_orientations=9, hog_pix_per_cell=8, hog_cell_per_block=2, hog_cells_per_step=2, hog_block_norm='L1', hog_channel='ALL', spatial_feat=True, hist_feat=True, hog_feat=True, heat_thresh=2, prev=None, prev_thresh=2)

Bases: object

bin_spatial(img, cspace0='BGR', visualize=False)

Compute binned color features

Parameters:

• img (numpy.ndarray) – The image for which we are computing features
• cspace0 (str) – The colorspace that img is currently in (e.g., ‘BGR’)
• visualize (bool) – Also return an image of the spatial color feature

Returns:

• numpy.ndarray – The stacked color features
• img (numpy.ndarray (if visualise=True)) – The color features image

color_hist(img, cspace0='BGR', visualize=False)

Compute color histogram features

Parameters:

• img (numpy.ndarray) – The image for which we are computing features
• cspace0 (str) – The colorspace that img is currently in
• visualize (bool) – Also return an image of the color histogram feature

Returns:

• numpy.ndarray – The concatenated histograms
• hists (list (if visualise=True)) – A list of the histograms

draw_bboxes(bbox_list, color=(0, 0, 255), thickness=6)

Draw boxes on the Frame object’s image

Parameters:

• bbox_list (list) – A list of ((x1, y1), (x2, y2)) coordinates for drawing boxes
• color (tuple) – The color that the boxes should be drawn
• thickness (int) – Thickness of lines that make up the rectangle

Returns:

draw_img – The image with boxes drawn on it

Return type:

numpy.ndarray

extract_features()

Extract HOG, spatial and/or color features from an image

This method is for extracting features from training images

See Frame.bin_spatial, Frame.color_hist, and Frame.get_hog_features

Returns:features – The concatenated features Return type:numpy.ndarray

find_cars(**kwargs)

Find cars using the heatmap

See Frame.get_heatmap

Parameters:**kwargs (dict) – Keyword arguments for Frame.get_bboxes / Frame.get_heatmap

get_bboxes(svc=None, X_scaler=None, scales=[1], min_rows=[None], max_rows=[None], cells_per_steps=[None])

Extract features from an image, apply the classifier, and generate self.heatmap

This method is for finding cars on a real world image (i.e., not a cropped training image)

See Frame.bin_spatial, Frame.color_hist, and Frame.get_hog_features

Parameters:

• svc (sklearn.svm.classes.LinearSVC) – SVM classifier
• X_scaler (sklearn.preprocessing.data.StandardScaler) – Feature scaler
• scales (list) – The scales (self.scale) at which to run the feature extraction and classifier
• min_rows (list) – A list of the minimum rows for cropping the image
• max_rows (list) – A list of the maximum rows for cropping the image
• cells_per_steps (list) – A list of the values for self.hog_cells_per_step

get_heatmap(**kwargs)

Extract features from an image, apply the classifier, and generate self.heatmap

See Frame.get_bboxes

Parameters:

• **kwargs (dict) – Keyword arguments for Frame.get_heatmap
• method is for finding cars on a real world image (i.e., not a cropped training image) (This) –
• Frame.bin_spatial, Frame.color_hist, and Frame.get_hog_features (See) –
• svc (sklearn.svm.classes.LinearSVC) – SVM classifier
• X_scaler (sklearn.preprocessing.data.StandardScaler) – Feature scaler
• scales (list) – The scales (self.scale) at which to run the feature extraction and classifier
• min_rows (list) – A list of the minimum rows for cropping the image
• max_rows (list) – A list of the maximum rows for cropping the image
• cells_per_steps (list) – A list of the values for self.hog_cells_per_step

get_hog_features(img, cspace0='BGR', visualize=False, feature_vector=False)

Extract a Histogram of Oriented Gradients (HOG) for the Frame object’s image

Parameters:

• img (numpy.ndarray) – The image for which we are computing features
• cspace0 (str) – The colorspace that img is currently in (e.g., ‘BGR’)
• visualize (bool) – Also return an image of the HOG
• feature_vector (bool) – Return the data as a feature vector by calling .ravel() on the result just before returning

Returns:

• features (list) – A list of HOG arrays (1 numpy array for each image channel)
• hog_image (list (if visualise=True)) – A list of visualisations of the HOG images (1 numpy array for each image channel)

get_windows()

Get a list of all the windows that will be searched for cars

Returns:bbox_list – A list with entries of the form (xy, xy_hog, xy_color), where each xy term is of the form ((x0, y0), (x1, y1)) and specifies a bounding box for the original image, the HOG-processed image, and the patch image used by Frame.color_hist and Frame.bin_spatial Return type:list

img

Return a Frame object’s rescaled associated image

Returns:The Frame object’s rescaled image Return type:numpy.ndarray

img0

Return a Frame object’s original associated image

Returns:The Frame object’s original image Return type:numpy.ndarray

load_img()

Load a Frame object’s associated image

project05.frame.convert_color(img, cspace=None, cspace0='BGR')

Convert img to colorspace cspace

Parameters:

• img (numpy.ndarray) – The image
• cspace (str, None) – The image will be converted to this colorspace (e.g., ‘BGR’)
• cspace0 (str) – The colorspace that img is currently in

Returns:

The image in the specified colorspace

Return type:

numpy.ndarray

project05.frame.heatmap_to_bboxes(heatmap)

Convert from a heatmap image to car bounding boxes

Parameters:heatmap (numpy.ndarray) – A heatmap image of car detections Returns:bbox_list – A list of ((x1, y1), (x2, y2)) coordinates for drawing boxes Return type:list