Python Library actspotter

The actspotter is a library / tensorflow model for detecting activities. It allows to classify body activities in images or videos. The package is limited to videos and images with only one person by design.

The following classes are available:

• none
• pull_up_up
• pull_up_down
• pull_up_none
• push_up_up
• push_up_down
• push_up_none
• sit_up_up
• sit_up_down
• sit_up_none

The package is currently in early development.

Future plans

Tensorflow model deployment will be integrated soon. Currently this package allows to classify push-ups, sit-ups and pull-ups. In future version kicks and others body activities will follow.

It is also planned to provide a signal processing layer that allows to easily detect connected activites and count them.

Another application will be to integrate with keyboard drivers so that activities could be used for controlling video games (e.g. by kicks).

Installation

Install this library in a virtualenv using pip. virtualenv is a tool to create isolated Python environments. The basic problem it addresses is one of dependencies and versions, and indirectly permissions.

With virtualenv, it's possible to install this library without needing system install permissions, and without clashing with the installed system dependencies.

Supported Python Versions

Python >= 3.6

Mac/Linux

pip install virtualenv
virtualenv <your-env>
source <your-env>/bin/activate
<your-env>/bin/pip install actspotter

Windows

pip install virtualenv
virtualenv <your-env>
<your-env>\Scripts\activate
<your-env>\Scripts\pip.exe install actspotter

Example Usage

Requirement: cv2 (opencv) installed.

Classification of images:

import cv2
import tensorflow as tf
from actspotter import ImageClassifier, classify_image_input_dimension, class_names

def _resize(frame, dim=classify_image_input_dimension):
    frame = cv2.resize(frame, dim, interpolation=cv2.INTER_AREA)
    return frame

def _to_tf_array(frame, dim=classify_image_input_dimension):
    frame = _resize(frame, dim)
    frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
    frame = tf.convert_to_tensor(frame, dtype=tf.float32)
    return frame

images = [
    to_tf_array(cv2.imread("test.jpg")),
]

print(class_names)
print(image_classifier.classify_images(images))

Classification of a video:

import cv2
import tensorflow as tf
from actspotter import VideoClassifier, classify_image_input_dimension

def _resize(frame, dim=classify_image_input_dimension):
    return frame

def _to_tf_array(frame, dim=classify_image_input_dimension):
    frame = _resize(frame, dim)
    frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
    frame = tf.convert_to_tensor(frame, dtype=tf.float32)
    return frame

cap = cv2.VideoCapture(0)

video_classifier = VideoClassifier(buffer_size=4)
video_classifier.start()

while cap.isOpened():
    ret, frame = cap.read()

    if ret == True:
        video_classifier.add_image(to_tf_array(frame))
        state = video_classifier.get_last_classification()
        print(state)

        frame = resize(frame, dim=(600, 600))
        cv2.putText(frame, f"{state}", (10, 40), 0, 2, 255)

        cv2.imshow("Frame", frame)

        waitkey = cv2.waitKey(25) & 0xFF

        if waitkey == ord("q"):
            break

video_classifier.exit()
cap.release()
cv2.destroyAllWindows()