A view of the application interface with a heat map overlayed on top
After seeing that there are many people that spends hours on hours analyzing repetative surveillance data, we wanted to make to lives of these individuals more productive and easier.
What it does
TrackyfAI is a video footage tool that detects all moving objects within a scene, whether it be a bike, a person, a car or truck, checks for unusual behaviour in the regions of the moving objects, and creates an importance contour of the scene over time.
How We Built It
We built our own custom OpenCV image processing algorithms to approximate contours to cropped images, and Tensorflow to unload our trained image classification models.
Challenges we an into
The machine learning aspect of this project deemed to be very challenging. In our case, we needed to estimate outcomes based on the relationship between different objects in the scene, and their individual changes over time. It is very difficult to find training data to create a statistical model for this procedure.
Accomplishments that we're proud of
We are proud of the different data visualization tools we were able to create for our application. From being able to view to view the movement history of a car driving in a scene, to viewing regions of importance in a current frame in the footage, we all thought these features were pretty cool.
What We learned
Windows and Unix don't work well together.