Object Tracking is a Computer Vision task used in many IVA (Intelligent Video Analysis) systems. This topic has an active research community, and every year many interesting ideas and algorithms pop up from GitHub and ArXiv. However, newcomers can find somewhat daunting to get started with Object Tracking research. If you find yourself in this situation or are just curious to know a little more about Object Tracking, then hopefully you will find this reading helpful.
At Wintics, we leverage Object Tracking algorithms to capture real-time data of any mobility infrastructure. This data falls mainly into two categories: (i) multimodal and directional counting and (ii) action recognition. The first type is useful for stating Origin-Destination matrices, which help understand how objects (i.e. persons and vehicles) come in and out the infrastructure. The second type can be more sophisticated: automatic incident detection, estimating the waiting time at a crossroad, detecting pick up and drop off of persons in airports, etc. All this information can be challenging to collect reliably in real-time. And often, it can be done by a human through watching videos from surveillance cameras, and this is where our solutions come in. Object Tracking plays a central role in almost all of our solutions, and this is why we do research on this field, to make our stack faster and more robust to any kind of camera and situation.
We were lucky to have Cindy Trinh as an intern last year, she worked on Object Tracking research among other subjects. At the end of her internship, she wrote a series of 3 medium posts on the subject of Object Tracking, which will help you acquire some context on how Object Tracking can be used to improve mobility and a good landscape of the recent research: