Deep neural networks have recently become increasingly used for a wide range of applications, (e.g., image and video processing). The demand for edge inference is growing, especially in the areas of relevance to the Internet-of-Things. Low-cost microcontrollers as edge devices are a promising solution for optimal application systems from several points of view such as: cost, power consumption, latency, or real-time execution. The implementation of these systems has become feasible due to the advanced development of hardware architectures and DSP capabilities, while the cost and power consumption have been maintained at a low level. The aim of the paper is to provide a literature review on the implementation of deep neural networks using ARM Cortex-M core-based low-cost microcontrollers. As an emerging research direction, there are a limited number of publications that address this topic at the moment. Therefore, the research papers that stand out have been analyzed in greater detail, to promote further interest of researchers to bring AI techniques to low power standard ARM Cortex-M microcontrollers. The article addresses a niche research domain. Despite the increasing interest manifested toward both (1) edge AI applications and (2) theoretical contributions in DNN optimization and compression, the number of existing publications dedicated to the current topic is rather limited. Therefore, a comprehensive literature survey using systematic mapping is not possible. The presentation focuses on systems that have shown increased efficiency in resource-constrained applications, as well as the predominant impediments that still hinder their implementation. The reader will take away the following concepts from this paper: (1) an overview of applications, DNN architectures, and results obtained using ARM Cortex-M core-based microcontrollers, (2) an overview of low-cost hardware devices and SW development solutions, and (3) understanding recent trends and opportunities.
Deep neural networks are currently applied in multiple domains, especially in the automotive industry. The main reason for this is related to the more complex challenges found in the field of signal processing, especially when the tasks involve image and video data types. Using conventional/statistical algorithms to deal with these high-complexity challenges is no longer a viable approach. Therefore, the involvement of artificial intelligence solutions like deep neural networks has significantly increased. In recent years, numerous architectures have been developed with the aim of maximizing performance. However, their size and computation requirements have increased at the same time. For this reason, special attention is currently being paid to the optimization of deep neural networks while trying to maintain (almost) the same performance. In this work, we aim to tackle the problem of eye gaze estimation considered within the automotive framework. Our proposal uses a knowledge distillation concept applied to a custom CNN architecture, called the teacher model. Based on this, several CNN student models are derived using layerwise and widthwise compression techniques. Furthermore, they are evaluated with respect to certain performance metrics, e.g. neural network size and inference time. In the experimental results, we propose certain compression methods which can address specific user requirements like model size, accuracy, and inference time. Finally, the student models are evaluated using an EdgeAI embedded device (STM32H747I-DISCO) in terms of accuracy, memory utilization, MACC complexity, and inference time. The combination of layerwise and widthwise compression results as the optimal method to derive student models with a good trade-off between the above-mentioned metrics. Using knowledge distillation, the accuracy can be improved by up to 9.5% over the conventional training procedure.