Edge AI Deployment: TinyML Models for Real-Time Object Detection on Resource-Constrained Devices
Download PDFAbstract
The proliferation of Internet of Things (IoT) devices has created a demand for on device intelligence, enabling real-time data processing at the edge. However, deploying deep learning models, particularly for computer vision tasks like object detection, on resource-constrained microcontrollers presents significant challenges due to their limited memory, computational power, and energy budgets. This chapter explores the domain of Tiny Machine Learning (TinyML) as a solution to this problem. We provide a comprehensive overview of the methodologies required to deploy lightweight object detection models on edge devices. The chapter details a complete workflow, from dataset selection and model training to advanced optimization techniques such as quantization, pruning, and knowledge distillation. We present a detailed analysis of the trade-offs between model accuracy, size, and inference latency for popular architectures like MobileNet and YOLO. Through simulated experiments, we evaluate the performance of these models on a typical microcontroller unit (MCU), analyzing key metrics including memory utilization, power consumption, and per class detection accuracy. The results demonstrate that with proper optimization, it is feasible to achieve real-time object detection on devices with less than MB of RAM, paving the way for a new generation of intelligent, battery-powered applications. The chapter concludes with a discussion of open challenges and future research directions in this rapidly evolving field.
References
- Alan Zilberman and Lindsey Ice. “Why computer occupations are behind strong STEM employment growth in the 2019–29 decade”. In: Computer 4.5,164.6 (2021), pp. 11–5.
- Syed Ali Raza Zaidi et al. “Unlocking edge intelligence through tiny machine learning (TinyML)”. In: IEEE Access 10 (2022), pp. 100867–100877.
- S Ren et al. “Towards real-time object detection with region proposal networks, Adv”. In: Neural Inf. Process 28 (2015).
- Joseph Redmon et al. “You only look once: Unified, real-time object detection”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2016, pp. 779–788.
- Wei Liu et al. “Ssd: Single shot multibox detector”. In: European conference on computer vision. Springer. 2016, pp. 21–37.
- Benoit Jacob et al. “Quantization and training of neural networks for efficient integer-arithmetic-only inference”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2018, pp. 2704–2713.
- Thomas Elsken, Jan Hendrik Metzen, and Frank Hutter. “Neural architecture search: A survey”. In: Journal of Machine Learning Research 20.55 (2019), pp. 1–21.
- Robert David et al. “TensorFlow lite micro: Embedded machine learning for tinyml systems”. In: Proceedings of machine learning and systems 3 (2021), pp. 800–811.
- Shawn Hymel et al. “Edge impulse: An mlops platform for tiny machine learning”. In: arXiv preprint arXiv:2212.03332 (2022).
- TY Lin et al. “Microsoft coco: Common objects in context, European Conf”. In: Computer Vision (Springer, Cham, 2014), pp. 740–755.
- Kiran Chand Ravi et al. “Ai-powered pancreas navigator: Delving into the depths of early pancreatic cancer diagnosis using advanced deep learning techniques”. In: 2023 9th International Conference on Smart Structures and Systems (ICSSS). IEEE. 2023, pp. 1–6.
- Andrew G Howard et al. “Mobilenets: Efficient convolutional neural networks for mobile vision applications”. In: arXiv preprint arXiv:1704.04861 (2017)
