HW-SW Optimization of DNNs for Privacy-Preserving People Counting on Low-Resolution Infrared Arrays
Risso et al. 2024 Design, Automation & Test in Europe Conference & Exhibition (DATE), Valencia, Spain, 2024, pp. 1-6, doi: 10.23919/DATE58400.2024.10546798.
Low-resolution infrared (IR) sensors facilitate people counting for occupancy and flow monitoring while ensuring privacy and reducing energy use. Deep Neural Networks (DNNs) efficiently process this data. This work, developed in the frame of the StorAIge project, introduces a full-stack optimization pipeline, from software to hardware, allowing low-resolution IR arrays to perform DNN inference at the edge. The proposed flow is able to improve significantly the state of the art on memory, code-size and energy reduction at iso- accuracy.
These results presented at the Design, Automation & Test in Europe Conference & Exhibition (DATE 2024) won the best paper award !
Abstract:
Low-resolution infrared (IR) array sensors enable people counting applications such as monitoring the occupancy of spaces and people flows while preserving privacy and minimizing energy consumption. Deep Neural Networks (DNNs) have been shown to be well-suited to process these sensor data in an accurate and efficient manner. Nevertheless, the space of DNNs’ archi-tectures is huge and its manual exploration is burdensome and often leads to sub-optimal solutions. To overcome this problem, in this work, we propose a highly automated full-stack optimization flow for DNNs that goes from neural architecture search, mixed-precision quantization, and post-processing, down to the realization of a new smart sensor prototype, including a Microcontroller with a customized instruction set. Integrating these cross-layer optimizations, we obtain a large set of Pareto-optimal solutions in the 3D-space of energy, memory, and accuracy. Deploying such solutions on our hardware platform, we improve the state-of-the-art achieving up to 4.2 x model size reduction, 23.8 x code size reduction, and 15.38 x energy reduction at iso-accuracy.
Please download the paper here: https://arxiv.org/pdf/2402.01226
More information on DATE conference: https://date24.date-conference.com/
