Saturday, September 07, 2019

Huawei Kirin 990 5G Camera Features

HuaweiCentral: Huawei presents its new mobile processor Kirin 990 5G at IFA 2019 in Berlin, Germany. One of its most impressive imaging features is the AI-based ability to determine the heart rate and breath rate just from a selfie camera video stream:



Another impressive feature is a real-time video segmentation:



More pictures form the company's IFA presentation:

Friday, September 06, 2019

DARPA Starts Curved IR Imagers Program

DARPA FOcal arrays for Curved Infrared Imagers (FOCII) program is created to expand upon the current commercial trend for visible sensor arrays by extending the capability to both large and medium format midwave (MWIR) and/or longwave (LWIR) infrared detectors. The program seeks to develop and demonstrate technologies for curving existing state-of-the-art large format, high performance IR FPAs to a small radius of curvature (ROC) to maximize performance, as well as curve smaller format FPAs to an extreme ROC to enable the smallest form factors possible while maintaining exquisite performance.

FOCII will address this challenge through two approaches to fabricating a curved FPA. The first involves curving existing state-of-the-art FPAs, while keeping the underlying design intact. The focus of the research will be on achieving significant performance improvements over existing, flat FPAs, with a target radius of curvature of 70mm. The fundamental challenge researchers will work to address within this approach is to mitigate the mechanical strain created by curving the FPGA, particularly in silicon, which is very brittle.

The second approach will focus on achieving an extreme ROC of 12.5 mm to enable a transformative reduction in the size and weight compared to current imagers. Unlike the first approach, researchers will explore possible modifications to the underlying design, including physical modifications to the silicon that could relieve or eliminate stress on the material and allow for creating the desired curvature in a smaller sized FPA. This approach will also require new methods to counter the effects of any modifications during image reconstruction in the underlying ROIC algorithm.


Thanks to TL for the link!

Thursday, September 05, 2019

LiDAR News: Lumotive, LeiShen, CoreDAR, Hitachi

GlobeNewswire: Lumotive, a Bill Gates-funded LiDAR startup, used Himax’s LCOS display with Lumotive’s patented Liquid Crystal Metasurfaces (LCMs) to improve the performance, reliability and cost of LiDAR systems. Other LiDAR sensors utilize MEMS mirrors or optical phased arrays. However, both of these approaches lack performance due to the small optical aperture of MEMS mirrors and the low efficiency of phased arrays. In a first for LiDAR, Lumotive leverages Himax’s unique, tailor-made LCOS process to convert semiconductor chips into dynamic displays that steer laser pulses based on the light-bending principles of metamaterials.

Lumotive’s LiDAR systems offer performance advantages, including a combination of:
  • Large optical aperture (25 x 25 mm) which delivers long range
  • 120-degree FoV with high angular resolution
  • Fast, random-access beam steering


Leishen Intelligent System presents its broad range of low-cost LiDARs. An automotive grade hybrid LiDAR CH16 3D is priced at $599 in quantities of 10,000:


Update: LeiShen kindly sent me their price list for small quantity purchases:


CoreDAR presents its tiny LiDAR concept:



Hitachi presents its view on LiDAR's role in smart city applications:

Wednesday, September 04, 2019

Samsung Exynos 980 Supports 108MP Camera

Samsung 5G 8nm Exynos 980 mobile processor supports up to 108MP camera:

"For advanced photography, the Exynos 980 delivers compelling camera performances with resolution support for up to 108-megapixels (Mp). The advanced image signal processor (ISP) supports up to five individual sensors and is able to process three concurrently for richer multi-camera experiences. Along with the NPU, the AI-powered camera is able to detect and understand scenes or objects, according to which the camera will then make optimal adjustments to its settings.

For an immersive multimedia experience, the Exynos 980’s multi-format codec (MFC) supports encoding and decoding of 4K UHD video at 120 frames per second (fps). HDR10+ support with dynamic mapping also offers more detailed and illuminant colors in video content.
"

sCMOS Sensors: Fairchild Imaging vs GPixel

Archiv.org paper "Evaluation of scientific CMOS sensors for sky survey applications" by S.Karpov, A.Bajat, A.Christov, M.Prouza from Czech Academy of Sciences compares Andor cameras based on Fairchild Imaging CIS2051 (Neo camera) and GPixel GSense400BSI (Marana camera) sCMOS sensors:

Scientific CMOS image sensors are a modern alternative for a typical CCD detectors, as they offer both low read-out noise, large sensitive area, and high frame rates. All these makes them promising devices for a modern wide-field sky surveys. However, the peculiarities of CMOS technology have to be properly taken into account when analyzing the data. In order to characterize these, we performed an extensive laboratory testing of Andor Marana sCMOS camera. Here we report its results, especially on the temporal stability and linearity, and compare it to the previous versions of Andor sCMOS cameras. We also present the results of an on-sky testing of this sensor connected to a wide-field lens, and discuss its applications for an astronomical sky surveys.

Tuesday, September 03, 2019

e2v Announces Fast Sensors

GlobeNewswire: Teledyne e2v announces its Flash CMOS sensor family, tailored for 3D laser profiling/displacement applications and high speed, high resolution inspection.

The new Flash sensors feature a 6μm CMOS global shutter pixel which effectively combines high resolution and fast frame rate. They are available in a 4k or 2k horizontal resolution, with respective frame rates of 1800fps and 1500fps (8 bits), and respective readout speeds of 61.4Gbps and 25.6Gbps (the best Gbps/price ratio in the market). The sensors come in a µPGA ceramic package fitting in standard optical formats, APS-like optics in the 4k and C-Mount in the 2k.

Yoann Lochardet, Marketing Manager for 3D at Teledyne e2v said, “We are very pleased to announce the release of the new Flash family of CMOS sensors which were developed after listening closely to the requirements of leading companies in the market. These new sensors feature a unique set of characteristics targeted at 3D laser triangulation applications including; high resolution, very high frame rate, very high readout speed, HDR capability and a large set of additional features. All these capabilities allow our customers to solve the most challenging application demands in 3D laser profiling/displacement such as quality control and 3D measurement.”

Evaluation Kits and samples of Flash 2K and Flash 4K are now available.

Omnivision Announces 2.2um Global Shutter Pixel with 40% QE at 940nm

PRNewswire: OmniVision announces the smallest-ever pixel size of 2.2um for a BSI, GS image sensor. The new OG01A sensor combines PureCel Plus-S pixel and Nyxel NIR technology to achieve QE of 40% at 940nm and 60% at 850nm.

The OG01A is well-suited to multiple machine-vision applications, including AR/VR headsets, drones, robots, and SLAM, as well as facial authentication in smartphones and other consumer electronics. This technology is also ideal for automotive in-cabin driver state monitoring and eye tracking.

"The OG01A has the industry's smallest global shutter pixel and provides the best NIR performance in a GS sensor," said Devang Patel, senior staff marketing manager for the security and emerging segments at OmniVision. "There is a growing need for global shutter technology to accurately capture images of moving objects, along with excellent NIR performance and small size, in camera applications such as AR/VR headsets, drones, robots and smartphones. The OG01A delivers the industry's best combination of features for these applications."

The 1.3MP OG01A sensor provides 1280x1024 resolution at 120 fps and 640x480 resolution at 240 fps in a compact 1/5 inch optical format. Samples are available now.

Monday, September 02, 2019

Face Recognition News

Panasonic will be providing additional automated facial recognition gates for use at a number of airports in Japan. Without the need for prior registration of biometric data, the system compares photographic data of the traveler’s face in the IC chip embedded in the passport with an image taken at the facial recognition gate to verify identity. According to the Japanese Ministry of Justice, approximately 80% of all Japanese travelers use Panasonic’s facial recognition gates.


Forbes: Hong Kong protesters destroying facial recognition towers in protest at a surveillance state. Meanwhile, Chinese facial recognition company Megvii had secured a $750m fundraise ahead of its IPO on Hong Kong stock exchange. Founded by three Tsinghua University graduates, Megvii is one of the major AI surveillance unicorns in China, alongside SenseTime, Yitu and CloudWalk. Reuters reported at the time that the funding raised Megvii's valuation to "slightly over $4 billion," as it prepared for its IPO in Hong Kong.

BusinessInsider, Petapixel: Hong Kong protesters are using laser pointers in an apparent attempt to blind facial-recognition cameras. At least one reporter posted a video with the damage made to his camera while shooting the protests:




My Modern Met: Avoiding facial recognition becomes an art. A piece of jewelry by Polish designer Ewa Nowak is claimed to confuse AI systems and helps a person to avoid the recognition:


South China Morning Post: China Pharmaceutical University in Nanjing is one of the first higher education institutions in the country to use facial recognition system to monitor the attendance and behavior of students.

“Besides attendance, the system installed in the classroom can provide surveillance of the students’ learning, such as whether they are listening to the lectures, how many times they raise their heads, and whether they are playing on their phones or falling asleep.

The school is taking action to cut down on students skipping class, leaving classes early, paying for a substitute to attend classes for them and not listening in class.
”

ON Semi Presents Full ISP Pipeline in CNN

ON Semi publishes Arxiv.org paper "Deep Camera: A Fully Convolutional Neural Network for Image Signal Processing" by Sivalogeswaran Ratnasingam.

"A conventional camera performs various signal processing steps sequentially to reconstruct an image from a raw Bayer image. When performing these processing in multiple stages the residual error from each stage accumulates in the image and degrades the quality of the final reconstructed image. In this paper, we present a fully convolutional neural network (CNN) to perform defect pixel correction, denoising, white balancing, exposure correction, demosaicing, color transform, and gamma encoding. To our knowledge, this is the first CNN trained end-to-end to perform the entire image signal processing pipeline in a camera. The neural network was trained using a large image database of raw Bayer images. Through extensive experiments, we show that the proposed CNN based image signal processing system performs better than the conventional signal processing pipelines that perform the processing sequentially."

Sunday, September 01, 2019

IISW 2019 Papers On-Line

2019 International Image Sensor Workshop (IISW) papers are available on-line at the International Image Sensor Society site. The workshop has been held on June 23-27, 2019
in Snowbird, Utah.