Compute and memory demands of state-of-the-art deep learning methods are still a shortcoming that must be addressed to make them useful at IoT end-nodes. In particular, recent results depict a hopeful prospect for image processing using Convolutional Neural Netwoks, CNNs, but the gap between software and hardware implementations is already considerable for IoT and mobile edge computing applications due to their high power consumption. This proposal performs low-power and real time deep learning-based multiple object visual tracking implemented on an NVIDIA Jetson TX2 development kit. It includes a camera and wireless connection capability and it is battery powered for mobile and outdoor applications. A collection of representative sequences captured with the on-board camera, dETRUSC video dataset, is used to exemplify the performance of the proposed algorithm and to facilitate benchmarking. The results in terms of power consumption and frame rate demonstrate the feasibility of deep learning algorithms on embedded platforms although more effort in the joint algorithm and hardware design of CNNs is needed.
Monocular depth estimation and image deblurring are two fundamental tasks in computer vision, given their crucial role in understanding 3D scenes. Performing any of them by relying on a single image is an ill-posed problem. The recent advances in the field of Deep Convolutional Neural Networks (DNNs) have revolutionized many tasks in computer vision, including depth estimation and image deblurring. When it comes to using defocused images, the depth estimation and the recovery of the All-in-Focus (Aif) image become related problems due to defocus physics. Despite this, most of the existing models treat them separately. There are, however, recent models that solve these problems simultaneously by concatenating two networks in a sequence to first estimate the depth or defocus map and then reconstruct the focused image based on it. We propose a DNN that solves the depth estimation and image deblurring in parallel. Our Two-headed Depth Estimation and Deblurring Network (2HDED:NET) extends a conventional Depth from Defocus (DFD) networks with a deblurring branch that shares the same encoder as the depth branch. The proposed method has been successfully tested on two benchmarks, one for indoor and the other for outdoor scenes: NYU-v2 and Make3D. Extensive experiments with 2HDED:NET on these benchmarks have demonstrated superior or close performances to those of the state-of-the-art models for depth estimation and image deblurring.
25 November 2021 in IEEE Transactions on Circuits and Systems II: Express Briefs
This brief presents an ultra-low power low-dropout (LDO) regulator with an experimental total quiescent current consumption of only 3.08 nA. The circuit is designed to operate with a load current in the range 0 - 11 mA. A novel adaptive biasing scheme based on a super source follower (SSF) structure is proposed, which measures the absolute voltage difference between the two inputs of the LDO’s error amplifier and modifies the biasing current accordingly. Thus, the transient response of the regulator is improved by counteracting the effect of using such a low bias current. The proposed LDO has been fabricated in a standard CMOS 180 nm process and the experimental characterization showed an outstanding performance in terms of maximum load current over quiescent current consumption ratio.
Unlike other noise sources, which can be reduced or eliminated by different signal processing techniques, shot noise is an ever-present noise component in any imaging system. In this paper, we present an in-depth study of the impact of shot noise on time-of-flight sensors in terms of the error introduced in the distance estimation. The paper addresses the effect of parameters, such as the size of the photosensor, the background and signal power or the integration time, and the resulting design trade-offs. The study is demonstrated with different numerical examples, which show that, in general, the phase shift determination technique with two background measurements approach is the most suitable for pixel arrays of large resolution.
19 October 2018 in IEEE Transactions on Power Electronics
This paper presents a Power Management Unit (PMU) powered by a 1 mm $^2$ solar cell on the same substrate to rise up the harvested voltage above 1.1 V. The on-chip solar cell and the PMU are fabricated in standard 0.18 μm CMOS technology achieving a form factor of 1.575 mm $^2$ . The PMU is able to start up from a harvested power of 2.38 nW without any external kick off or control signal. The PMU features a continuous and two-dimensional Maximum Power Point Tracking (MPPT) working in open-loop mode to handle a harvested power range from nW to $\mu$ W, by modifying both the charge pump topology and the switching frequency. The MPPT is based on four voltage level detectors that define five working regions depending on the illumination and on a self-tuning reference current for a fine adjustment of the switching frequency. The chip also includes an auxiliary charge pump to generate the voltage level necessary for the control circuit, implemented as a Pelliconi charge pump of 8 stages with NMOS transistors in Pwell as diodes. A Dickson charge pump with transmission gates as switches and with variable gain and capacitance per stage is also designed as the main charge pump. Finally, two relaxation oscillators are implemented to drive both charge pumps. This paper is accompanied by a video file demonstrating the PMU operation by powering an off-chip NAND gate.