Real-time Object Detection Performance Analysis Using YOLOv7 on Edge Devices

Abstract

Real-time object detection in images is one of the most important areas in computer vision and finds applications in several fields, such as security systems, protection, independent vehicles, and robotics. Many of these applications need to use edge hardware platforms, and it is vital to know the performance of the object detector on these hardware platforms before developing the system. Therefore, in this work, we executed performance benchmark tests of the YOLOv7-tiny model for real-time object detection using a camera and three embedded hardware platforms: Raspberry Pi 4B, Jetson Nano, and Jetson Xavier NX. We tested and analyzed the NVIDIA platforms and their different power modes. The Raspberry Pi 4B achieved an average of 0.9 FPS. The Jetson Xavier NX achieved 30 FPS, the maximum possible FPS rate, in three power modes. In the tests, it was possible to notice that the maximum CPU clock of the Jetson Xavier NX impacts the FPS rate more than the GPU clock itself. The Jetson Nano achieved 7.4 and 5.2 FPS in its two power consumption modes.