LIVE COVID Streamer and Alarming System Problem Statement: There is a lot of problem, for Governments to find the current stats of COVID infected people, also despite of lock down lots of people are still getting out and violating public safety rules. Also it is important to know infected cases in each region so that security forces can be deployed accordingly
Suggestion: There must be a system through which governments can find rational stats of infected people and also set security forces where this system suggests that social distancing is being voilated.
Solution: I am designing a solution to this, with following tasks.
Detect Humans. Detect number of humans. Calculate distance between them. Create alarm if they violate social distancing. Detect if detected human has flu like symptoms or not, so that infected people can be counted per area/region/city/Country.
Human Detection and Distance Calculation: I’ve used ResNet to first detect human with online available trained weights, then I count number of humans in it and count distance between their coords with Euclidean distance, also I fixed the size of image before passing it to Neural Network in order to get steady distances always. I’ve always prepared it to work on Live Stream. It gives Alarm when two or more persons come close to each by certain distance, this threshold of distance is calculated with lots of test images. This is the repo of ResNet I used for this task. “https://github.com/cftang0827/pedestrian_recognition”
Flu Symptoms Detection: I used this dataset “https://web.bii.a-star.edu.sg/~chengli/FluRecognition.htm” in which I found videos of classes included with flu-like symptoms, I first extract images from it by noticing action timings in videos and train ResNet with Transfer Learning with these images.
I started all this only before two days (26th of March, 2020), thus accuracy of trained models is bit compromised. But all codes are complete and represent my complete idea. This works on live stream too. This is the repo I used for this task. “https://www.kaggle.com/suniliitb96/tutorial-keras-transfer-learning-with-resnet50”
Try It out
computer-vision, flu-detection, human-counter, human-detection, human-distance