Our group is passionate about computer vision and with the prompt "New wave of tech," we thought it would be a good opportunity to learn more about computer vision and while combining it with problems that have risen this year.
What it does
Our program can determine the number of people in a certain area covered by a camera. It will count the number of people, calculate the percent capacity, and send the information to the store owner. With this, store owners can cut costs from employees having to be stationed at the entrance of stores. It will also keep employees safer by avoiding high traffic areas.
How we built it
We used Yolov3 with their trained weights as a framework. OpenCV webcam functions were used to feed frames into Yolov3 and some Python functions are added to calculate percentages and keep track of capacity. Along-side, Twilio API was used to send SMS messages to an external device from the program for notifications.
Challenges we ran into
We tried to use the GoogleVision framework at first and got it to work for the most part; however, we had problems with the datatypes it outputted and the general framework was a little harder to implement with OpenCV and some other ideas we had. Yolov3 was a much easier approach and because of their pre-trained tiny weights, it allowed our program to boost its fps up. We will be presenting our Yolov3 version.
Accomplishments that we're proud of and what we learned
We are proud that we have learned the basic framework for both Google vision and Yolov3. Having the challenge of doing both helped us learn the similarities between frameworks and we now have a higher understanding of the topic in general.
What's next for Twice Vision
TwiceVision at the moment is only useful by enforcing covid regulations; however, after the pandemic, the general program will still serve as a useful tool. Because it is trainable, it can be adaptable to do many other tasks. An example of this could be keeping a record of stock or keeping track of suspicious behavior in stores potentially allowing the store owner to stop a crime from taking place. CCTV has an article about implementing this, where they used their footage of suspicious behaviour to train a program. If TwiceVision can train on other datasets, it will be a useful tool for store owners and allow employees to be assigned to other tasks.