This thesis explores the viability of using constrained embedded devices in object detection systems, where close-to-real-time system responsiveness is required. To achieve low latency, the computationally intensive tasks, not suitable for low-cost embedded devices, were offloaded to a remote processing computer over a Wi-Fi network. A proof-of-concept system was developed where a Raspberry Pi captured and pre-processed images. The data was then transmitted to a more powerful computer, which used its CPU to run an object detection inference algorithm (YOLOv11m). Finally, detection data was sent back to the Raspberry Pi for usage. Results showed that the Raspberry Pi maintained low CPU loads during tests, which confirmed its ability to perform additional tasks as a smart client with low latency. Furthermore, tests were conducted to identify drop-offs in detection performance given decreased image resolution and image frame rate. If reliable object detection is required, results suggest minimal image compression, with a frame rate of at least 1 image per second for the camera capture area used in this study. In scenarios where the Wi-Fi connection is unstable or when minimizing the data usage is of importance, image compression is advised. This study identified near-original detection performance even when images were downscaled to approximately 60% of their original resolution (from 640x130 pixels to 384x78 pixels).