Best Projects 2020 features all the nominated entries submitted ‘Submit your Project’ category. All the entries consist of innovative projects run by existing enterprises in the form of businesses, NGOs or informal programs.
JUGAAD for a "Robust Social Distancing Monitoring System"
Published on August 17, 2020 um 13:05
Principal Data Scientist
Type of Enterprise
Year the ngo or company was founded
Explain your project in details:
"Andrew Ng" described COVID19 social distancing alert system here: https://landing.ai/landing-ai-creates-an-ai-tool-to-help-customers-monitor-social-distancing-in-the-workplace/ However, if the camera has distortions or the persons are at different planes then it won't work properly. Also computation cost is more for his algorithm than my JUGAAD (https://en.wikipedia.org/wiki/Jugaad). My approach is as follow: suppose widths of two persons appearing in an image are w1 & w2 pixels respectively. Also suppose 1.2 meter horizontal distance (minimum separation for COVID19 safety) appearing in image is d pixels. I found with little experiment that [(w1+w2)/2d] is near about 0.8 for most of the typical CCTV cameras. Hence this knowledge helps me in finding the distance between two persons (person detection I am doing with YOLO Deep Learning Algorithm) without doing perspective-correction or camera-distortion-correction. As width is taken into consideration so the algorithm works well even for persons appearing partially (vertically) in the image or persons at different planes. No need to calibrate CCTV cameras- in general. Little fine-tuning on the value (0.8) might be required, & then it works good for almost all CCTV cameras & in roughly all situations. Few more "false positives" would be there; that's why this is a "Jugaad".
Impact of your enterprise on sustainable development
TVS Motor Company is a leading 2-wheeler & 3-wheeler manufacturer with head office in India. I have implemented & deployed this solution to prevent spreading of COVID19 across various offices/plants of my employer (covering nearly 30 cameras). This is working superb there. Please watch the generated alerts in the given link below. 4 different camera results are shown. Red lines indicated the violation. Also please note that I have set-up a threshold for minimum width to ignore far people. Link: https://drive.google.com/drive/folders/131P928vQmHApWGeonw9MgHZDhYhJebKd?usp=sharing
Sustainability and future plans
This project was for our company's internal purpose only to prevent spreading COVID19 in office premises. We can think about other future use-cases of this as: To detect presence of unauthorised person in company premises; To detect unwanted behaviour among employees inside office premises etc.
Your profile as an entrepreneur
I am 41 years old. I am having education from premier institutes with 14 years of multi-domain experiences in Image Processing, Computer Vision, Deep Learning & related areas; I have worked in domains like education, mining, automotive, healthcare, retail, banking, automobile etc. Here is my LinkedIn profile link: https://www.linkedin.com/in/sujoy-kumar-goswami/
Sample Output Results