In order to build a reliable, scalable, and competitive product for Lot Spot, we needed a state of the art machine learning algorithm to track and collect data on vehicles. While I did not build the system (I do not know machine learning code), I worked closely with the dev team to create a system architecture that could handle the tech stack I wanted.
We needed the system to run on scalable and reliable hardware. I stress tested many single board computers and concluded on the Nvidia Tx2. The hardware also needed a modem, cameras, and other components. When the hardware was chosen, we needed to design a system that could support the accuracy we were looking for while operating on an off-grid power supply. The design included: hardware, libraries, models, system tools, power management systems, and communication methods.
The end result was a machine learning computer vision algorithm that ran off .8W and was 99.8% accurate (industry leading).
Client: Lot Spot Inc.