Performance Comparison of Edge Devices for Deep Learning-Based Object Detection and Classification

In the era of rapid technological advancement, object detection has become a crucial component in various applications, ranging from facial recognition to autonomous vehicles. This study aims to compare the performance of three popular devices commonly used for object detection, namely Raspberry Pi, Coral, and Jetson Nano. The evaluation was conducted using various TensorFlow Lite and TensorRT models to measure execution times for classification and detection tasks. The dataset used consists of thousands of images and videos that is collected personally and from publicly accessible datasets such as Caltech256, ensuring the uniqueness and diversity of the testing data. It shows that Jetson Nano is the fastest machine in out tested devices with execution speed of 2.6 ms for classification tasks, this is followed by the google coral TPU with 50 ms, and raspberry pi taking the last position at 500 ms in this task category, in the object detection category our study shows that Jetson nano still outperforms other machines with execution time of around 7 ms, while the coral TPU hovers around 100 ms, and the raspberry pi takes 500 ms to 1000 ms depending on the model type. The main advantage of this study is the high validity of the results obtained through a large and varied dataset collected personally, providing deep insights into the performance of each device for classification and object detection tasks.