Data Processing¶
Person entity¶
-
class
source.entities.person.Person(heatmap=None, offsets=None, rescale=(1, 1), threshold=0.7, path_txt=None, show_head=False, output_stride=32)¶ Person detected in a frame. With this class a person entity can be extracted from an image. All the keypoints are extracted and keypoints saved.
- Returns:
- Person:
-
get_height()¶ keypoints: 15: LEFT FOOT, 16: RIGHT FOOT, 0: NOSE. :return:
-
get_width()¶ keypoints: 5: LEFT SHOULDER, 6: RIGHT SHOULDER. :return:
-
infer_lc_keypoints(prev_person)¶ This function will be used when creating the frame groups. It takes the Person of the previous frame and uses its keypoits to infer the keypoints of “self”. If the keypoint has low confidence, the position of the keypoint relative to the neck is used in the current person
- Args:
- prev_person (Person): The person from the current frame
-
infer_point(index, prev_person)¶ Use the position of the neck and the same keypoint from the previous frame to infer this one. Same confidence is applied
- Args:
- index (int): Index of the keypoint to infer prev_person (Person): Person from the previous frame
-
is_valid_first()¶ This function determines if the frame should be considered for training. Before it was embedded inside the pipeline of DataProcessor. Now it’s a function, so conditions can be changed based on performance.
- Returns:
- bool: True if is valid
-
class
source.entities.person_frames.PersonMovement(list_persons, times_v=10, joints_remove=(13, 14, 15, 16), model='LSTM')¶ Extracts coordinates from a list of persons. The coordinates extracted are the input for all the models that predict actions.
- It has two main functions:
- get_vector: Get positions of keypoints and joint speeds. Used for Non LSTM models
- get_vector: Get positions of keypoints. Input for LSTM model.
- Returns:
- PersonMovement:
-
get_vector(times_v, joints_remove)¶ Get coordinates vector from a series of frames.
- Args:
- times_v (int): Times the body velocity is repeated in the resulting vector. joints_remove (tuple): Joints that will be removed and not used in the final vector
- Returns:
- ndarray: Flattened vector of [x + v * times_v + v] dimensions. Where x and v are the flattened vectors of joints positions and velocities.
-
get_vector_lstm(joints_remove)¶ Get coordinates vector (only positions) from a series of frames.
- Args:
- joints_remove (tuple): Joints that will be removed and not used in the final vector
- Returns:
- ndarray: Flattened vector of [x] dimensions. Where x is the flattened vector of joints positions and velocities.
Feed Forward Model¶
Firefox Controller¶
-
class
source.controller.Controller(classes, initial_url=None, driver_path=None, time_rotation=0.5, coordinates=None)¶ Class to control the movement in Google Street View. It only works with Firefox. If there is no guess about where to start (initial_url or coordinates) it starts in Zaragoza (Spain).
- Returns:
- Controller:
-
perform_action(i_action)¶ Perform an action given its index within self.classes
- Args:
- i_action (int): index of classes list
-
perform_action_name(action)¶ Perform action given its name.
- Args:
- action (str): Name of the action to perform. As it uses a mapper, it must be one of [‘walk’, ‘stand’, ‘left’, ‘right’]
-
class
source.controller.DistanceMeter(starting_url=None, starting_coords=None, units='km')¶ This class is used to keep track of the distance made during a walk. It retrieves coordinates from a google maps url using regex and calculates distance walked.
-
static
distance_from_coords(coords1, coords2)¶ Gets two pairs of coordinates in decimal format and calculates distance between points
- Args:
- coords1 (list): Coordinates in decimal format (lat, long) coords2 (list): Coordinates in decimal format (lat, long)
- Returns:
- float: Meters between points
-
retrieve_coords(url)¶ From a Google Maps URL it obtains the coordinates
- Args:
- url (str): URL from which coordinates have to be extracted
- Returns:
- list: Coordinates in decimal format (lat, long)
-
update_distance(url)¶ self.distance keeps record of all the distance made in the walk. This function updates this value.
- Args:
- url (str): URL of the new step
-
static
