SDDetector 0.1.0

lightweight video detection

SocialDistance

Keep safe social distance is considered as an effective way of avoiding spreading of coronavirus. Our SocialDistance module is a lightweight package which provides an implementation of utlizing deep learning models for monitoring safe social distance.

Demo

Watch the demo video

Dataset

We use the video clip collected from OXFORD TOWN CENTRE dataset and made the above demo video.

Supported Models

We have tested our model using Faster-RCNN, YOLO-v3 and SSD, based on the performance of each model, we have chosen YOLO-v3 as our default model

All our models are pretrained models from Gluno CV Tookit

Installation

You may be able to obtain the latest version our model from:

pip install -i https://test.pypi.org/simple/ SocialDistance==0.1
pip install gluoncv
pip install mxnet-cu101

Usage

After Successfully installed SocialDistance, you can use it for detection by:

from SocialDistance.utils.Run import Detect
detect = Detect()
#you may want to give an image as input to check the validity of bird-eye view transformation
detect(image)

If no arguments is given, our model will run using the default data collected from 'OXFORD TOWN CENTRE' dataset, otherwise you may want to specify arguments expicitly:

from SocialDistance.utils.Run import Detect
detect = Detect(video_path, video_save_path, keypoints, keypoints_birds_eye_view, actual_length, actual_width, pretrained_models)
#you may want to give an image as input to check the validity of bird-eye view transformation
detect(image)

Parameters

• video_path: input path of video
• video_save_path: output path of labelled video
• keypoints: selected key points from first frame of the input video
• keypoints_birds_eye_view: mapping location of keypoints on the bird-eye view image
• actual_length: actual length in real-world
• actual_width: actual width in real-world
• pretrained_models: selected pretrained models