This is a dashboard for monitoring one meter social distancing for outdoor areas using computer vision. This can be implemented realtime on CPU thanks to OpenVino model optimisation. To track objects (people), SORT algorithm with Kalmann Filter (tracker) and Hungarian algorithm (data association) has been used. Pre-trained openvino people detection model (MobileNetV2 SSD) has been used for detection. Perspective transformation has been used to get the bird's eye view to enhance the distance measurement accuracy. OpenCV is used for image processing.
Demo: https://www.youtube.com/watch?v=joDZVKSOvhM
Install OpenVino 2021.4
Create a new conda environment. If you dont have conda installed download miniconda
conda create -n sdt python=3.8 Clone this repository to your computer and navigate to the directory.
Activate new enviroment
conda activate std Install all the libraries used
pip install -r requirements.txt Initialize OpenVino
source /opt/intel/openvino_2021/bin/setupvars.sh Run the script run_sort.py
usage: run_sort.py [-h] [-i INPUT_FILE] -m MODEL_PATH [-t THRESHOLD]
[-o OUTPUT_FILE] [-c CAMERA] [--save] [--find_homography]
[--no_bbox] [--slow]
Run SORT
optional arguments:
-h, --help show this help message and exit
-i INPUT_FILE, --input_file INPUT_FILE
input videos file path name
-m MODEL_PATH, --model_path MODEL_PATH
path to the model
-t THRESHOLD, --threshold THRESHOLD
threshold for detections
-o OUTPUT_FILE, --output_file OUTPUT_FILE
output video file path name
-c CAMERA, --camera CAMERA
camera stream index
--save whether to save the video
--find_homography if the transformation matrix is not available
--no_bbox circles will be drawn instead of bounding boxes
--slow reduce the fps of the video if too highWith Bounding Boxes for Visualization
