sideseeing-tools 0.6.5

A set of tools to load, preprocess and analyze data collected through the MultiSensor Data Collection App

SideSeeing Tools

SideSeeing Tools is a collection of scripts designed to load, preprocess, and analyze data gathered through the MultiSensor Data Collection App.

Installation

pip install sideseeing-tools

General Usage

Create a dataset

from sideseeing_tools import sideseeing

ds = sideseeing.SideSeeingDS(root_dir='/home/user/my-project')

# Available iterators
#   .instances  // Tip: dictionary of instances (key=name, value=SideSeeingInstance)
#   .iterator   // Tip: for i in ds.iterator: i.name

# Available attributes and methods
#   .metadata() // Tip: generates and prints the dataset metadata
#   .size       // Tip: shows the number of instances  
#   .sensors    // Tip: a dictionary containing the names of the available sensors

Get a random sample from the dataset

my_sample = ds.instance

Check the available sensors by instance

ds.sensors
# This command will produce output like this:

{
    # A key representing the number of available axes
    'sensors1': {
        # A key representing the sensor name
        'lps22h barometer sensor': {
            # Keys representing the instances where the sensor data is found
            'FhdFastest#S10e-2024-08-01-10-42-43-354',
            'FhdGame#S10e-2024-08-01-10-25-08-383',
            'FhdNormal#S10e-2024-08-01-10-02-18-947',
            'FhdUi#S10e-2024-08-01-10-13-50-369'
        },
        'tcs3407 uncalibrated lux sensor': {
            'FhdFastest#S10e-2024-08-01-10-42-43-354',
            ...
        },
        ...
    },
    'sensors3': {
        'ak09918c magnetic field sensor': {...},
        'bmi160_accelerometer accelerometer non-wakeup': {
            'FhdFastest#Mia3-2024-08-01-10-42-44-639',
            'FhdNormal#Mia3-2024-08-01-10-02-22-118',
            ...
        },
        ...
    },
    'sensors6': {
        ...
    }
}

Get accelerometer data from the sample

my_sample = ds.instances['FhdNormal#Mia3-2024-08-01-10-02-22-118']
my_sample_accel_data = my_sample.sensors3['bmi160_accelerometer accelerometer non-wakeup']
my_sample_accel_data

	Datetime UTC	x	y	z	Time (s)
0	2024-03-21 19:33:01.550000	9.34247	-0.270545	3.10767	0
1	2024-03-21 19:33:01.561000	9.51725	-0.347159	3.00233	0.011
2	2024-03-21 19:33:01.571000	9.46458	-0.407014	2.81079	0.021
3	2024-03-21 19:33:01.581000	9.35205	-0.395043	2.79164	0.031
4	2024-03-21 19:33:01.590000	9.36402	-0.263362	2.77488	0.04

Extract a snippet from a sample (video and sensor data)

my_sample.extract_snippet(
    start_time=2,                        # Start time of the snippet (in seconds)
    end_time=17,                         # End time of the snippet (in seconds)
    output_dir='/home/user/snippet_2_17' # Directory to save the extracted snippet
)

Running the command extract_snippet will generate one file for the video (with audio), one file for consumption data, one file for GPS data, and one file for each sensor type (single-axis, three-axis, three-axis uncalibrated) present in the instance. See an illustrative example in the following file tree.

home/
├─ user/
│  ├─ snippet_2_17/
│  │  ├─ consumption.2_17.csv
│  │  ├─ gps.2_17.csv
│  │  ├─ sensors.one.2_17.csv
│  │  ├─ sensors.three.2_17.csv
│  │  ├─ sensors.three.uncalibrated.2_17.csv
│  │  ├─ video.2_17.mp4

Extract only video snippets

It is possible to extract a snippet from a video (and only the video) using the extract_video_snippet function.

from sideseeing_tools import media

# Extract a 15-second snippet from the video, beginning at the 3-second mark and ending at the 18-second mark
media.extract_video_snippet(
    source_path=my_sample.video,    # Path to the input mp4 file
    start_second=3,                 # Start time of the snippet (in seconds)
    end_second=18,                  # End time of the snippet (in seconds)
    output_path='my_snippet.mp4'    # Path to save the extracted snippet
)

Iterate over the samples

for i in ds.iterator:
    print(i.name, i.video)

Create a plotter

from sideseeing_tools import plot

plotter = plot.SideSeeingPlotter(ds, taxonomy='/home/user/my-project/taxonomy.csv')

# Available methods:
#   .generate_video_for_sensor()
#   .plot_dataset_cities()
#   .plot_dataset_map()
#   .plot_dataset_tags_matrix()
#   .plot_dataset_tags()
#   .plot_instance_audio()
#   .plot_instance_map()
#   .plot_instance_sensors3_and_audio()
#   .plot_instance_video_frames_at_times()
#   .plot_instance_video_frames()
#   .plot_sensor()
#   .plot_sensors()

Frame Extraction Methods

The following methods provide flexible and efficient ways to extract frames from video files, either saving them to disk or returning them in memory for immediate use in your applications. While using the methods below, you can opt for saving the frames to disk or returning them in memory.

extract_frames_at_times

This method extracts frames from a video file at the specified time points and saves them as image files.

Parameters:

• source_path (str): Path to the source video file.
• frame_times (list): List of time points (in seconds) at which frames should be extracted.
• target_dir (str, optional): Path to the target directory where frames will be saved. If None, frames will be returned in memory.
• prefix (str, optional): Prefix to be added to the frame file names.
• left_zeros (int, optional): Number of left-padded zeros in the frame file names.

Returns:

• list: List of paths to the extracted frames or list of frames in memory if target_dir is None.

extract_frames_at_positions

This method extracts frames from a video file at the specified frame positions and saves them as image files.

Parameters:

• source_path (str): Path to the source video file.
• frame_positions (list): List of frame positions at which frames should be extracted.
• target_dir (str, optional): Path to the target directory where frames will be saved. If None, frames will be returned in memory.
• prefix (str, optional): Prefix to be added to the frame file names.
• left_zeros (int, optional): Number of left-padded zeros in the frame file names.

Returns:

• list: List of paths to the extracted frames or list of frames in memory if target_dir is None.

extract_frames_timespan

This method extracts frames from a video file within a specified time span and saves them as image files.

Parameters:

• source_path (str): Path to the source video file.
• start_time (float): The start time (in seconds) of the time span from which frames should be extracted.
• end_time (float): The end time (in seconds) of the time span from which frames should be extracted.
• target_dir (str, optional): Path to the target directory where frames will be saved. If None, frames will be returned in memory.
• step (int, optional): The rate at which frames are extracted.
• prefix (str, optional): Prefix to be added to the frame file names.
• left_zeros (int, optional): Number of left-padded zeros in the frame file names.

Returns:

• list: List of paths to the extracted frames or list of frames in memory if target_dir is None.

extract_frames_positionspan

This method extracts frames from a video file within a specified position span and saves them as image files.

Parameters:

• source_path (str): Path to the source video file.
• start_frame (int): The start frame position from which frames should be extracted.
• end_frame (int): The end frame position from which frames should be extracted.
• target_dir (str, optional): Path to the target directory where frames will be saved. If None, frames will be returned in memory.
• step (int, optional): The rate at which frames are extracted.
• prefix (str, optional): Prefix to be added to the frame file names.
• left_zeros (int, optional): Number of left-padded zeros in the frame file names.

Returns:

• list: List of paths to the extracted frames or list of frames in memory if target_dir is None.

extract_frames

This method extracts frames from a video file at a specified frame rate and saves them as image files.

Parameters:

• source_path (str): Path to the source video file.
• target_dir (str, optional): Path to the target directory where frames will be saved. If None, frames will be returned in memory.
• step (int, optional): The rate at which frames are extracted.
• prefix (str, optional): Prefix to be added to the frame file names.
• left_zeros (int, optional): Number of left-padded zeros in the frame file names.

Returns:

• list: List of paths to the extracted frames or list of frames in memory if target_dir is None.

Example Usage

from sideseeing_tools import media

# Extract frames at specific times and save to disk
media.extract_frames_at_times(
    source_path=video_path,
    frame_times=[1.0, 2.0, 3.0],
    target_dir='output',
    prefix='frame_',
    left_zeros=5
)

# Extract frames at specific times and return in memory
frames = media.extract_frames_at_times(
    source_path=video_path,
    frame_times=[1.0, 2.0, 3.0]
)

# Extract frames at specific positions and save to disk
media.extract_frames_at_positions(
    source_path=video_path,
    frame_positions=[30, 60, 90],
    target_dir='output',
    prefix='frame_',
    left_zeros=5
)

# Extract frames at specific positions and return in memory
frames = media.extract_frames_at_positions(
    source_path=video_path,
    frame_positions=[30, 60, 90]
)

# Extract frames within a timespan and save to disk
media.extract_frames_timespan(
    source_path=video_path,
    start_time=10.0,
    end_time=20.0,
    target_dir='output',
    step=30,
    prefix='frame_',
    left_zeros=5
)

# Extract frames within a timespan and return in memory
frames = media.extract_frames_timespan(
    source_path=video_path,
    start_time=10.0,
    end_time=20.0,
    step=30
)

# Extract frames within a position span and save to disk
media.extract_frames_positionspan(
    source_path=video_path,
    start_frame=300,
    end_frame=600,
    target_dir='output',
    step=30,
    prefix='frame_',
    left_zeros=5
)

# Extract frames within a position span and return in memory
frames = media.extract_frames_positionspan(
    source_path=video_path,
    start_frame=300,
    end_frame=600,
    step=30
)

# Extract frames at a specified frame rate and save to disk
media.extract_frames(
    source_path=video_path,
    target_dir='output',
    step=30,
    prefix='frame_',
    left_zeros=5
)

# Extract frames at a specified frame rate and return in memory
frames = media.extract_frames(
    source_path=video_path,
    step=30
)

Additional tips

We suggest implementing the following folder structure: create a directory named data to contain all recordings. By doing so, when instantiating the SideSeeingDataset, a metadata.csv file will be generated in the root directory. Here is the command to instantiate a dataset:

ds = sideseeing.SideSeeingDS('/home/user/my-project', subdir='data', name='MyDataset')

And here is the suggested folder structure:

my-project/
├─ data/
│  ├─ place01/
│  │  ├─ route01/
│  │  │  ├─ consumption.csv
│  │  │  ├─ gps.csv
│  │  │  ├─ metadata.json
│  │  │  ├─ sensors.one.csv
│  │  │  ├─ sensors.three.csv
│  │  │  ├─ sensors.three.uncalibrated.csv
│  │  │  ├─ video.gif
│  │  │  ├─ video.mp4
│  │  │  ├─ video.wav
│  │  ├─ route02/
│  ├─ place02/
│  ├─ place03/
├─ metadata.csv
├─ taxonomy.csv

Sensor data specification before SideSeeing conversion

The following data outlines the specifications of sensor content before SideSeeing conversion, i.e., when accessing them directly through the files generated by the MultSensor Data Collection tool.

File consumption.csv

datetime_utc	battery_microamperes
2024-03-21T19:38:04.961Z	-1431
2024-03-21T19:38:05.961Z	-1011
2024-03-21T19:38:06.961Z	-2216

File gps.csv

datetime_utc	gps_interval	accuracy	latitude	longitude
2024-03-21T19:38:10.309Z	15	16.0	-23.5645676	-46.7395994
2024-03-21T19:38:38.033Z	15	57.639	-23.5645617	-46.739602
2024-03-21T19:38:54.120Z	15	26.611	-23.5645528	-46.7396658

File sensors.one.csv

timestamp_nano	datetime_utc	name	axis_x	accuracy
0	2024-03-13T13:40:27.243Z	Palm Proximity Sensor	5.0	3
712657771915658	2024-03-21T19:38:05.015Z	TCS3407 Uncalibrated lux Sensor	1810.0	3
712657931915658	2024-03-21T19:38:05.174Z	TCS3407 Uncalibrated lux Sensor	1812.0	3

File sensors.three.csv

timestamp_nano	datetime_utc	name	axis_x	axis_y	axis_z	accuracy
712657652031560	2024-03-21T19:38:04.895Z	LSM6DSO Acceleration Sensor	9.603442	-0.10295067	3.9959226	3
712657673895658	2024-03-21T19:38:04.916Z	LSM6DSO Acceleration Sensor	9.823709	-0.38067806	3.9097314	3
712657652031560	2024-03-21T19:38:04.895Z	LSM6DSO Gyroscope Sensor	0.113544576	0.42852196	0.083306745	3

File sensors.three.uncalibrated.csv

timestamp_nano	datetime_utc	name	axis_x	axis_y	axis_z	delta_x	delta_y	delta_z	accuracy
712657851915658	2024-03-21T19:38:05.094Z	Uncalibrated Magnetic Sensor	268.56	-9.54	-230.45999	255.48	-2.28	-227.94	3
712657852615658	2024-03-21T19:38:05.096Z	Gyroscope sensor UnCalibrated	0.044593163	-0.13439035	0.07086037	-0.003009122	-0.016193425	-0.0026664268	3
712657862615658	2024-03-21T19:38:05.105Z	Gyroscope sensor UnCalibrated	0.042760566	-0.05009095	0.100792766	-0.003009122	-0.016193425	-0.0026664268	3
712657874678965	2024-03-21T19:38:05.118Z	Uncalibrated Magnetic Sensor	268.62	-8.639999	-231.48	255.48	-2.28	-227.94	3
712657872615658	2024-03-21T19:38:05.116Z	Gyroscope sensor UnCalibrated	0.064751714	-0.007330383	0.118507855	-0.003009122	-0.016193425	-0.0026664268	3

List of SideSeeingInstance attributes/methods

Attribute or method	Description
geolocation_points	Dictionary containing geolocation data, including latitude and longitude coordinates representing geographical points.
geolocation_center	Latitude and longitude coordinates representing the geographic center of a specific area.
audio	Path to the audio file associated with the collected data.
gif	Path to the GIF file associated with the collected data.
video	Path to the video file associated with the collected data.
sensors1	Dictionary containing data from one-axis sensors.
sensors3	Dictionary containing data from three-axis sensors.
sensors6	Dictionary containing data from six-axis sensors, including uncalibrated data.
label	List of categories and tags representing the taxonomy of sidewalks.
video_start_time	Start time of the video associated with the collected data.
video_stop_time	Stop time of the video associated with the collected data.
extract_snippet	Extracting a snippet from the sample (video, sensor, gps and consumption data).

Author

Rafael J P Damaceno

About us

The SideSeeing Project aims to develop methods based on Computer Vision and Machine Learning for Urban Informatics applications. Our goal is to devise strategies for obtaining and analyzing data related to urban accessibility. Take a look at our website.