Ambarella Inc.


Ambarella, Inc. is a fabless semiconductor design company, focusing on low-power, high-definition and Ultra HD video compression, image processing, and computer vision processors. Ambarella's products are used in a wide variety of human and computer vision applications, including video security, advanced driver assistance systems, electronic mirror, drive recorder, driver and in-cabin monitoring, autonomous driving, and robotics applications. Ambarella's system on chips are designed to deliver a combination of video compression, image processing, and computer vision performance with low-power operation to enable cameras to extract data from high-resolution video streams.

History

Ambarella was founded in 2004 by Feng-Ming Wang and Les Kohn with the goal of developing high-definition H.264 video encoders for the professional broadcast market. Soon after, Ambarella applied this same technology to consumer video and security IP camera markets, focusing on the development of low-power, compression-efficient chips capable of producing high-quality imagery in challenging lighting and high-motion environments. Over the next decade, Ambarella chips were featured in a number of notable consumer camera products, including the GoPro Hero, the Dropcam by Nest, and the DJI Phantom series of drones.
In July 2015 Ambarella acquired VisLab, a pioneer in perception systems and autonomous vehicle research founded by Professor Alberto Broggi. VisLab has a history of developing computer vision and intelligent control systems for automotive and commercial applications, including ADAS and autonomous vehicles. Subsequent generations of Ambarella SoCs incorporated VisLab perception technologies at the hardware level, with the goal of targeting automotive OEM camera designs across all of SAE’s six levels of driving automation.

Technologies

Ambarella develops intelligent embedded processors for a range of camera markets—including security, wearable, drone, sports/action, and automotive—with an emphasis on several core technologies.
Dedicated Hardware Architecture for Computer Vision
Known as CVflow®, Ambarella’s computer vision architecture includes a dedicated hardware engine programmed with a high-level algorithm description, allowing convolutional neural networks trained with industry-standard tools to be mapped onto CVflow-based chips. This approach represents a marked difference from that of general-purpose architectures, such as those offered by GPUs and CPUs, which are popular alternatives for computer vision processing.
Stereovision
While Ambarella chips are capable of monocular processing, stereovision is a key focus area for the company. Using cameras with multiple lenses and image sensors, Ambarella’s stereovision processing enables the capture of three-dimensional imagery via the simulation of binocular vision, making it possible to detect generic obstacles that a perception system hasn’t been trained to recognize. This generic 3D obstacle detection can be valuable when deployed in partial or fully autonomous robotics applications where atypical objects may be encountered.
Edge Design
In contrast to cloud-dependent hardware technologies, Ambarella’s technologies are designed for use in edge applications, where a significant share of the processing is performed locally, onboard the device. To meet the demands of processing at the edge, Ambarella products prioritize low power consumption, reduced physical footprint, and extended operating temperature ranges.
Image Processing
Ambarella processors include specialized digital image processing hardware to convert raw image sensor data into color-corrected imagery, removing noise to improve quality in various lighting conditions, as well as performing lens distortion correction, digital image stabilization, and high dynamic range processing.
The resulting image is either
encoded for streaming or saving as a file,
processed by computer vision algorithms to sense the surrounding environment,
presented to the user on a visual display.
Video encoding
Ambarella SoCs use data compression techniques to encode video streams into popular industry-standard coding formats, including H.264 and H.265.

Applications

Ambarella processors are designed specifically for automotive, security, consumer, and industrial/robotics camera applications.
Automotive
Ambarella automotive SoCs target both aftermarket and OEM camera designs. Relevant applications include simple drive recorders, front ADAS cameras, driver and in-cabin monitoring systems, electronic mirrors, 360° surround view systems with parking assistance, and stereovision cameras for vehicular autonomy. Key automotive features include lane departure warning, lane keeping, forward collision warning, auto emergency braking, intelligent headlight control/high beam assist, speed assist, auto parking assist, blind spot detection, target tracking, generic obstacle detection, terrain modeling, curb/barrier detection, and sensor fusion.
Security
The company’s professional security and home monitoring camera applications place an emphasis on object detection/classification, face detection and recognition, person/pet/vehicle detection, license plate recognition, behavioral analysis, and high-resolution image processing in low-light and high-contrast environments. Target camera designs include indoor/outdoor cameras, multi-sensor cameras, intelligent transport systems and traffic cameras, retail cameras, wire-free cameras, and smart video doorbells.
Consumer
Building on its history in powering mass-market consumer video devices, Ambarella continues to target sports/action cameras, VR cameras, drones, and wearable cameras. Ambarella consumer processors prioritize high-resolution image processing, video compression, low-power operation, and computer vision features such as smart video editing, target-of-interest tracking, and augmented reality.
Industrial and Robotics
Ambarella computer vision processors are designed to enable a variety of intelligent robotics applications, including automated guided vehicles, consumer robots, and industrial/machine vision solutions. Key features of interest include low-latency computer vision performance, neural network processing, stereovision capabilities, and high-resolution/high-framerate video analysis.

Awards and Recognition