Color-Coding Traffic Paths Using an Object Tracking Algorithm on a Low-Power GPU

Tracking moving objects with code isn’t a new concept. There are many computer vision libraries that have functions that can track changes in position very accurately. This allows for computers to be able to provide data about situations that aren’t able to be observed in a reasonable amount of time. For example, tracking hundreds of moving cars over a day would take a lot of time if done by hand, but with code, one can get that data quicker. This thesis aims to provide a clear, simple, and effective application to track moving objects in a given video, trace their paths, and color-code these paths to see which ones are the most congested. This is to provide an efficient and deployable algorithm to track moving objects. This research was in collaboration with Moog Inc, an aerospace and defense company, to develop an algorithm that would analyze a video of a parking lot and determine the empty parking spaces and the common traffic paths that cars take while in a parking lot. Moog Inc. provides an Optimized Development Environment (ODE) to develop the application. Since the hardware is efficient on power and has a small form factor, the applications that are run on it are very easily deployable and portable, which makes it useful for any environment. The process of tracking cars in a video is somewhat straightforward as well. It consists of filtering the video, drawing rectangles around each region (car), tracing their paths (movements) and applying a heatmap to that path. Since it isn’t too computationally intensive, it can work well on the ODE. Since the ODE is small and has a portable form factor, this algorithm can be deployed fairly easily. The heatmap generation was effective in showing the densities of certain paths that cars traveled through. There are also various colormaps that can be used, to provide a clearer idea of the paths. There were attempts to optimize this algorithm by processing every other frame instead, but ultimately the tradeoff between efficiency and accuracy was deemed to be unfavorable. There were still some limitations that this approach had, as initially the algorithm would draw paths between areas that weren’t traversed by cars. While this was fixed in the final result, there are still some slight inaccuracies within the roads. There are also ethical concerns with the use of this software, as Moog Inc. does a lot of work in defense and this software could be used in wartime scenarios. However, this software can be applied to various other scenarios like tracking wildlife in an area to study their habits, or tracking particles to see their density in a given environment. Since the algorithm is ran on a low-powered environment, it can be deployed and tested in many different scenarios without being costly.