Object detection in edge computing challenges and solutions for resource-limited devices

The challenge of balancing accuracy and efficiency has driven the broad application of MobileNet and DenseNet detection in IoT scenarios while achieving high accuracy with low latency using an embedded device with limited computational power and memory resources remains a significant challenge. This paper investigates the object detection capabilities of two mainstream models, MobileNet and DenseNet, in edge computing environments. In particular, we investigate the computational resource demands of all models while deployed on embedded devices—a parameter size, memory usage, and inference latency—with very narrow hardware capabilities. We give experimental results about the parameter size of both models, memory usage, as well as detection accuracies tested on edge devices. Results have also shed some light on trade-offs among model complexity, accuracy, and resource efficiency that offer clear indications about every model's suitability for actual deployment on any resource-constrained device in an edge computing perspective. The research will also be useful, from a practical point of view, to investigate the deployability of the MobileNet and DenseNet models on IoT devices and contribute to developing efficient and effective solutions aimed at object detection and suited to resource-limited environments.