Computer Vision is becoming widely used for a myriad of purposes, e.g. people counting and tracking. To execute this application in real-time, a relatively complex algorithm processes intensive data streams to identify people in a visual scenario. Although such algorithms frequently run in powerful servers on the Cloud, it is also common that they have to run in local commodity computers with limited capacity. In this work we used the Multi-Camera Multi-Target algorithm of the recent OpenVINOTM toolkit to detect and track people in small retail stores. We ran the algorithm in a common personal computer and analyzed the variation of its performance for a set of different relevant scenarios and algorithm configurations, providing insights into how these affect the algorithm performance and computational cost. In the tested scenarios, the most influential factor was the number of people in the scene. The average frame processing time observed varied around 200 ms.