Pasar al contenido principal

INTELIGENCIA EMBEBIDA EN SENSORES Y SISTEMAS DE VISION DE BAJO CONSUMO Y OPERACION DE LARGO TERMINO

PID2021-128009OB-C31

•
Nombre agencia financiadora Agencia Estatal de Investigación
Acrónimo agencia financiadora AEI
Programa Programa Estatal para Impulsar la Investigación Científico-Técnica y su Transferencia
Subprograma Subprograma Estatal de Generación de Conocimiento
Convocatoria Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Año convocatoria 2021
Plan Estatal 2021-2023 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario AGENCIA ESTATAL CONSEJO SUPERIOR DE INVESTIGACIONES CIENTIFICAS (CSIC)

Resultados relacionados

Resultados totales (Incluyendo duplicados): 9
Encontrada(s) 1 página(s)

DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Artículo científico (JournalArticle)

Reliable and efficient integration of AI into camera traps for smart wildlife monitoring based on continual learning

  • Velasco-Montero, Delia
  • Fernández-Berni, J.
  • Carmona-Galán, R.
  • Sanglas, Ariadna
  • Palomares, Francisco
In this paper, we comprehensively report on an efficient approach for the integration of artificial intelligence (AI) processing pipelines in camera traps for smart on-site wildlife monitoring. Our work covers hardware, software, and algorithmics. We have built two prototypes of smart camera trap on a maximum bill of materials of 100$. We have also built two datasets, made publicly available, comprising over 17 k images, many of them notably challenging even for humans. Leveraging our broad expertise on embedded systems, specialized software libraries and toolchains, and AI techniques such as transfer learning, explainable AI, and, most importantly, continual learning, we achieve more reliable inference on-site - specifically 10 % higher F1-score - than MegaDetector run off-site on a desktop computer. The paper includes many practical details on system realization and on-site training in addition to a vast set of lab and experimental results., This work was supported by project SEMIoTICS (PID2021-128009OB-C31), funded by MCIN/AEI/10.13039/501100011033 and EU “ERDF: A way of making Europe”; by project ULTIMATE (TED2021-131835B-I00), funded by MCIN/AEI/10.13039/501100011033 and by the European Union NextGenerationEU/PRTR; by project SUMHAL (LIFEWATCH-2019-09-CSIC-4, POPE 2014-2020), funded by Ministry of Science and Innovation of Spain through European Regional Development Fund; by the Ministry for Digital Transformation and Public Function through grant USECHIP (TSI-069100-2023-001) of PERTE Chip Chair Program, funded by European Union – Next Generation EU; and by project SIMMRE, “Plan Complementario de I+D+i en el área de Biodiversidad (PCBIO)” funded by the European Union within the framework of the Recovery, Transformation and Resilience Plan - NextGenerationEU, by the Spanish Ministry of Science, Innovation and Universities and by the Regional Government of Andalucia., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Artículo científico (JournalArticle)

Single-Shot Auto-Exposure and High Dynamic Range Imaging Based on Asynchronous Operation of Pixel Array Circuitry

  • Lamouaraa-Sedlackova, Yassine
  • Fernández-Berni, J.
  • Carmona-Galán, R.
This paper introduces an image formation technique that realizes automatic adaptation to the illumination conditions during image capture. This adaptation takes place asynchronously in a single shot, requiring neither external control nor iterations on the exposure setting. These characteristics are especially suitable for applications such as visual robot navigation, where illumination may abruptly change and immediate decisions must be made. The equations that model the proposed technique show exposure-time independence. Moreover, each pixel adapts itself according to the ratio between local and global illuminations. The dependence on this ratio—which is a relative rather than absolute magnitude—along with the aforementioned independence on exposure time means that, theoretically, any illumination scenario can be represented within an arbitrary signal range. We also address the CMOS implementation of this technique. An extensive set of electrical simulations confirms its auto-exposure and high-dynamic-range encoding capabilities. In particular, we simulated the capture of a 118-dB scene on a (Formula presented.) -px array over five orders of magnitude of ambient illumination. Corner and mismatch simulations reveal strong robustness of the proposed circuitry against fabrication non-idealities. Finally, although we aim to applications in which large image resolution is not a critical specification, strategies to minimize the impact of the required focal-plane circuit elements on the pixel pitch are examined., This work has been supported by Project SEMIoTICS (PID2021-128009OB-C31), funded by MCIN/AEI/10.13039/501100011033 and EU “ERDF: A way of making Europe,” and by Project ULTIMATE (TED2021-131835B-I00), funded by MCIN/AEI/10.13039/501100011033 and by the European Union NextGenerationEU/PRTR., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Artículo científico (JournalArticle)

Compressive Image Encoder Architecture for In-Sensor Feature-Space Sampling

  • Khan, Amir
  • Fernández-Berni, J.
  • Carmona-Galán, R.
An architecture for the on-chip implementation of a compressive image encoder is presented. It is 100% compatible with standard CMOS image sensor architectures. It does not interfere with normal CMOS image sensor operation. No significant image degradation is expected. A stream of compressed vectors containing feature-space samples is provided in parallel with the full raw image data. The major technical challenges posed by the additional in-sensor circuitry are: introducing negligible disturbances on image capture and maintaining image resolution and quality. Concerning the compressive sampling measurement strategy, it is generated on-chip, without requiring storage or communication of the measurement matrix. A new pseudo-diagonal random ternary measurement matrix is proposed that generates compressed vectors that require only 6-bit ADC resolution to render accurate classification results. In addition, the proposed architecture is flexible to implement other measurement strategies. The proposed encoder architecture is implemented in a 110nm CMOS image sensor technology., Spanish Ministry of Science, Innovation and Universities (MCIN/AEI /10.13039/501100011033) through the SEMIoTICs Project under Grant PID2021-128009OB-C31; in part by the EU “European Regional Development Fund (ERDF): A way of making Europe” funded by MCIN/AEI/10s.13039/501100011033 under Grant PRE2019-088478; in part by the EU “European Social Fund (ESF): Investing in your future” and the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)— Project Mem2CNN under Project 441957207; in part by the DFG Priority Program SPP 2262 MemrisTec under Project 422738993; and in part by the Secretary of State for Telecommunications and Digital Infrastructure, Ministry for Digital Transformation and Civil Service and by European Union–NextGenerationEU/PRTR through the Chair in Microelectronics of the University of Seville (USECHIP) Project under Grant TSI-069100-2023-001., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

Attention-driven PCM-based In-Memory Computing for Smart Vision Systems

  • Adnan Haidar, Amir Khan, Vasileios Ntinas, Jorge Fernández-Berni, Ricardo Carmona-Galán, Ronald Tetzlaff
This work is funded by BMBF via DAAD project 57616814 (SECAI, School of Embedded Composite AI), and by MCIN/AEI via project SEMIoTICS (PID2021-128009OB-C31), and also by DFG through project Mem2CNN (#441957207) of SPP MemrisTec (#2262). Additional funding is provided by the Ministry for Digital Transformation through grant USECHIP (TSI-069100-2023-001) of PERTE Chip, funded by the Next Generation-EU., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

A Novel 10T SRAM Bit-Cell with High Static Noise Margin and Low Power Consumption for Binary In-Memory Computing

  • Khosravi, H.
  • Carmona-Galán, R.
  • Fernández-Berni, J.
  • Kandalaft, N.
In this paper, a novel 10T SRAM bit-cell with a high read/write static noise margin and low power dissipation is presented to be utilized in binary in-memory computing. The proposed SRAM bit-cell employs a conventional 6T SRAM along with four extra transistors to decouple read and write operation paths and improve the stability of the bit-cell. The two transistors M7 and M8 in the proposed structure act as switches to enable binary operation of the dot product in IMC mode. To conduct a comparative analysis, the various SRAM bit-cell designs were simulated using the Cadence design framework in the 65 nm technology. The proposed SRAM bit-cell not only presents a higher read stability value among the considered cells but also has a proper read delay value. The read noise margin of the proposed circuit is 1.91× as compared to a conventional 6T SRAM bit cell. Furthermore, the read energy of the proposed 10T SRAM has the lowest value of 28 fJ at a supply voltage of 1V compared to all the considered SRAM bit cells., R. Carmona-Galán and J. Fernández-Berni have been supported by Project PID2021-128009OB-C31 funded by MCIN/ AEI /10.13039/501100011033/ and ERDF, a way to make Europe., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

Hardware-Efficient Random-Modulation ΣΔ ADC for Per-Column CS Generation in Vision Sensor

  • Khan, Amir
  • Fernández-Berni, J.
  • Carmona-Galán, R.
Compressive sensing (CS) encodes multiple projections of a signal. If the signal is an image, each projection is a linear combination of all pixels, which requires a high-resolution analog-to-digital converter (ADC). Σ Δ ADCs have been widely used to digitize compressed samples with practical resolution. In this study, we present the circuit-level implementation of a hardware-efficient random-modulation first-order incremental Σ ΔADC. This ADC can perform random modulation of input signals by ±1 while simultaneously quantizing and averaging the signals. The random modulation takes place directly at the decimation filter without requiring additional circuitry, in contrast to state-of-the-art works. The ADC has been designed for per-column CS realization in a vision sensor featuring a pixel pitch of 4 μm. However, the scheme is general and can be employed in any other relevant CS application. All circuits have been designed using a 110 nm CMOS image sensor technology., This work has been supported by Project SEMIOTICS (PID2021- 128009OB-C31), funded by MCIN/AEI /10.13039/501100011033 and EU “ERDF: A way of making Europe”, and by Grant PRE2019-088478, funded by MCIN/AEI /10.13039/501100011033 and EU “ESF: Investing in your future”., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

An Analog-to-Information Architecture for Single-Chip Sensor-Processor Inference System

  • Khan, Amir
  • Fernández-Berni, J.
  • Carmona-Galán, R.
Compressed sensing (CS) establishes that signals fulfilling certain conditions can be sampled at a much lower rate than that defined by the Nyquist-Shannon sampling theorem while still keeping most information. Therefore, the monolithic implementation of CS is notably interesting for different applications. By its very nature, CS encodes global information of the sampled signal and hence compressed samples can be used for inference purposes. Thus, a particular application of CS is the implementation of inference in the compressed domain, i.e., without reconstructing the compressed signal. In this paper, we present an analog-to-information CS-based architecture for on-chip inference. The performance of the proposed architecture has been evaluated using facial recognition as a case study for different scenarios with varying lighting, expression, and occlusion conditions. Different mixed-signal hardware building blocks are also presented. These building blocks have been designed in the 110 nm LFoundry technology., This work has been supported by Project SEMIOTICS (PID2021- 128009OB-C31), funded by MCIN/AEI /10.13039/501100011033 and EU “ERDF: A way of making Europe”, and by Grant PRE2019-088478, funded by MCIN/AEI /10.13039/501100011033 and EU “ESF: Investing in your future”., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

Accurate Face Recognition on Highly Compressed Samples

  • Khan, Amir
  • Fernández-Berni, J.
  • Carmona-Galán, R.
Compressive sensing is an emerging field for low-dimensional data acquisition. Samples are acquired in the compressed domain and utilized for signal reconstruction or as input features for a classifier. In this work, hardware-aware face recognition using compressed samples was investigated. A linear support vector machine (SVM) classifier was exploited with compressed samples as input features; Faces can be reliably recognized with high average accuracy (up to 99%). To assess the robustness of the proposed scheme, three image datasets covering different facial and illumination conditions were analyzed. Random (binary) and structured (Haar-transform-based) measurement matrices were employed for generating compressed samples. For one of the datasets, Extended Yale B, and using a random binary measurement matrix, the proposed scheme achieved 82% accuracy from as few as 15 compressed samples, which means a 1/20480 sensing ratio. Accuracy and compression are also remarkably high with respect to the state-of-the-art for the other two datasets., This work has been supported by Project SEMIOTICS (PID2021- 128009OB-C31), funded by MCIN/AEI/10.13039/501100011033 and EU ”ERDF: A way of making Europe”, and by Grant PRE2019-088478, funded by MCIN/AEI/10.13039/501100011033 and EU ”ESF: Investing in your future”., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Otros (Other)

Hardware-Efficient Random-Modulation ΣΔ ADC for Per-Column CS Generation in Vision Sensor, Hardware-Efficient Random Modulation ΣΔ ADC for Vision Sensor Based Per-Column CS Generation

  • Khan, Amir
  • Fernández-Berni, J.
  • Carmona-Galán, R.
Título del postprint: Hardware-Efficient Random Modulation Σ∆ ADC for Vision Sensor Based Per-Column CS Generation. -- Trabajo presentado en la 30th IEEE International Conference on Electronics, Circuits and Systems (ICECS), celebrada 4-7 Dec. 2023, Istambul, Turkey, Compressive sensing (CS) encodes multiple projections of a signal. If the signal is an image, each projection is a linear combination of all pixels, which requires a high-resolution analog-to-digital converter (ADC). Σ Δ ADCs have been widely used to digitize compressed samples with practical resolution. In this study, we present the circuit-level implementation of a hardware-efficient random-modulation first-order incremental Σ ΔADC. This ADC can perform random modulation of input signals by ±1 while simultaneously quantizing and averaging the signals. The random modulation takes place directly at the decimation filter without requiring additional circuitry, in contrast to state-of-the-art works. The ADC has been designed for per-column CS realization in a vision sensor featuring a pixel pitch of 4 μm. However, the scheme is general and can be employed in any other relevant CS application. All circuits have been designed using a 110 nm CMOS image sensor technology., This work has been supported by Project SEMIOTICS (PID2021-128009OB-C31), funded by MCIN/AEI /10.13039/501100011033 and EU “ERDF: A way of making Europe”, and by Grant PRE2019-088478, funded by MCIN/AEI /10.13039/501100011033 and EU “ESF: Investing in your future”., Peer reviewed