53 A Datacompressive 15B275b Loggradient Qvga Image Sensor With Multiscale Readout For Alwayson Object Detection

• doi: 10.1109/ISSCC.2019.8662436
• title: 5.3 A Data-Compressive 1.5b/2.75b Log-Gradient QVGA Image Sensor with Multi-Scale Readout for Always-On Object Detection
• publisher: IEEE
• isbn: 978-1-5386-8532-7
• issn: 0193-6530
• rank: 102
• access_type: LOCKED
• content_type: Conferences
• abstract: Histograms of Oriented Gradients (HOG) are attractive features for object detection in embedded vision applications, as they provide a good trade-off between complexity and detection accuracy. A custom 8b CMOS imager that computes these features on-chip consumes only 52pJ/pixel [1]. However, a complete system also requires a backend detection algorithm, which consumes 940pJ/pixel in an optimized implementation [2]. As shown in the system study of [3], this imbalance mostly stems from the large amount of data seen by the detector. To remedy this issue, the work of [3] studies a feature-extraction approach that aggressively log-quantizes the data, thereby eliminating unnecessary illumination-related bits from the histograms. The custom log-gradient image sensor described in this paper demonstrates this concept in CMOS. It consumes 127pJ/pixel and offers two log-gradient modes (1.5b and 2.75b) along with multi-scale readout. With several algorithmic enhancements enabled by the log gradients as described in [3], the resulting HOG feature compression ratios are 25× (1.5b) and 9.5× (2.75b) when compared to a standard 8b image (without image pyramid). Using 1.5b log gradients, [3] conservatively estimates a 3.3× reduction in backend detection energy for a deformable parts model (DPM) based detector.
• article_number: 8662436
• pdf_url: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8662436
• html_url: https://ieeexplore.ieee.org/document/8662436/
• abstract_url: https://ieeexplore.ieee.org/document/8662436/
• publication_title: 2019 IEEE International Solid-State Circuits Conference - (ISSCC)
• conference_location: San Francisco, CA, USA
• conference_dates: 17-21 Feb. 2019
• publication_number: 8656625
• is_number: 8662285
• publication_year: 2019
• publication_date: 17-21 Feb. 2019
• start_page: 98
• end_page: 100
• citing_paper_count: 6
• citing_patent_count: 0
• download_count: 1366
• insert_date: 20190307
• index_terms:
  • ieee_terms:
    • Lighting
    • Feature extraction
    • Object detection
    • Detectors
    • Capacitors
    • Image sensors
    • Histograms
  • dynamic_index_terms:
    • Object Detection
    • Image Sensor
    • Camera Sensor
    • Large Amount Of Data
    • Detection Accuracy
    • Linear Gradient
    • Logarithmic Scale
    • Log-scale
    • Log Scale
    • Standard Imaging
    • Complete System
    • Average Precision
    • Energy Reduction
    • Vertical Gradient
    • Large Dynamic Range
    • High Frame Rate
    • Low Light Levels
    • Histogram Of Oriented Gradients
    • Image Pyramid
    • Image Pyramids
    • Bias Point
    • Biased Point
    • Selective Switching
• isbn_formats:
  • format: Print on Demand(PoD) ISBN, value: 978-1-5386-8532-7, isbnType: New-2005
  • format: Electronic ISBN, value: 978-1-5386-8531-0, isbnType: New-2005
• authors:
  • Author Name: Christopher Young
    Affiliation: Stanford University, Stanford, CA
    Author URL: https://ieeexplore.ieee.org/author/37086372228
    ID: 37086372228
    Order: 1
    Author Affiliations:
    • Stanford University, Stanford, CA
  • Author Name: Alex Omid-Zohoor
    Affiliation: Stanford University, Stanford, CA
    Author URL: https://ieeexplore.ieee.org/author/37086372239
    ID: 37086372239
    Order: 2
    Author Affiliations:
    • Stanford University, Stanford, CA
  • Author Name: Pedram Lajevardi
    Affiliation: Robert Bosch, Sunnyvale, CA
    Author URL: https://ieeexplore.ieee.org/author/38106129000
    ID: 38106129000
    Order: 3
    Author Affiliations:
    • Robert Bosch, Sunnyvale, CA
  • Author Name: Boris Murmann
    Affiliation: Stanford University, Stanford, CA
    Author URL: https://ieeexplore.ieee.org/author/37300093100
    ID: 37300093100
    Order: 4
    Author Affiliations:
    • Stanford University, Stanford, CA

Image Sensor

• sensor_type: CMOS
• resolution: QVGA (320x240)
• dynamic_range: Not explicitly mentioned
• pixel_size: 5 µm
• dark_current: Not explicitly mentioned

Optical Data

• focal_length: Not explicitly mentioned
• aperture: Not explicitly mentioned
• field_of_view: Not explicitly mentioned
• distortion: Not explicitly mentioned

Performance Metrics

• frame_rate: Not explicitly mentioned
• signal_to_noise_ratio: Not explicitly mentioned
• sensitivity: Not explicitly mentioned
• shutter_speed: Not explicitly mentioned
• power_consumption: 127 pJ/pixel
• noise: Not explicitly mentioned

Applications & Benefits

• cell_imaging: Used in object detection within embedded vision applications; this sensor computes features such as Histograms of Oriented Gradients (HOG) on-chip.
• benefits: Data compression ratios of 25× (1.5b) and 9.5× (2.75b) when compared to a standard 8b image; reduced backend detection energy.

Supporting Organizations

• supported_by: Bosch RTC; funded in part by Systems on Nanoscale Information fabrics (SONIC), MARCO and DARPA.

Manuscript Details

• publication_date: 17-21 Feb. 2019

Relevancy Score

• score: 9
• missing_fields:
  • dynamic_range
  • dark_current
  • focal_length
  • aperture
  • field_of_view
  • distortion
  • frame_rate
  • signal_to_noise_ratio
  • sensitivity
  • shutter_speed
  • noise

Processed JSON Filename

request_4687a6a5-3e8e-4401-8ad5-4515a4cbf3b1-53_a_datacompressive_15b275b_loggradient_qvga_image_sensor_with_multiscale_readout_for_alwayson_object_detection.json