Ambarella chip

CVflow™ – Advanced Computer Vision Architecture

The CVflow™ architecture is based on a deep understanding of core computer vision algorithms.

Compared with general-purpose CPUs and GPUs, CVflow™ uses a fundamentally different approach that includes a flexible CV hardware engine programmed with a high-level algorithm description. This allows our architecture to scale performance to trillions of operations per second with very low power and without limiting flexibility. Customers can efficiently map their CNN networks trained with industry-standard tools such as Caffe and TensorFlow to run on CVflow™-based chips.

  • CV1 - 4K Stereovision Processor with CVflow™ Computer Vision Architecture
  • CV22 - 4K SoC with CVflow™ Computer Vision Architecture
Faces in the crowd image

Computer Vision

Our technologies incorporate advanced computer vision algorithms into a variety of video applications including smart IP cameras, automotive video solutions (ADAS, smart electronic mirrors, intelligent surround view systems, and self-driving vehicles), and next-generation drones.

  • Stereovision processing detects generic objects without training
  • Deep neural network (DNN) performance
  • Terrain modeling, traffic light detection, 3D free space detection, lane detection, and CNN classification for vehicles, pedestrians, and bicycles/motorcycles
  • Path planning, target tracking, obstacle avoidance, multi-point navigation, enhanced safety shield, and GPS-denied precise visual localization
  • Full vehicle autonomy
  • CVflow™ architecture
Skydiver with helmet cam

Low-Power, High-Quality HD and Ultra HD Video

Ambarella tapes out multiple chips every year at the most advanced process node available, allowing the delivery of products that are both high performance and extremely low power.

  • Exceptionally efficient multi-core processor architecture
  • Innovative algorithms achieve a level of video quality that has become the industry’s performance benchmark
  • Low power consumption enables the development of small form-factor cameras
Nightview camera

High-Performance Image Processing

In any digital camera, the raw video data originating from the imaging sensor requires intelligent pre-processing prior to video encoding. Ambarella’s groundbreaking imaging algorithms — combined with ultra-high-speed pixel processing — enhance dynamic range, eliminate noise and deliver outstanding color fidelity under the most challenging lighting conditions.

  • MCTF filtering
  • High-ISO video processing
  • High Dynamic Range (HDR) functionality resolves detail in high-contrast environments
  • Increased compression efficiency — bandwidth isn't wasted on encoding unwanted noise
Displays showing HD quality video

State-of-the-Art Encoding

Ambarella has become an established leader in HD and Ultra-HD video encoding solutions. As our next generation of high-performance chips are released, we are continuing to define new standards of H.264 and H.265 encoding excellence to enable the HD and Ultra-HD cameras of the future.

  • Highly efficient encoding provides outstanding video quality with lowest bitrates for HD and Ultra HD applications
  • “Smart AVC” streaming architecture enables extremely low bit rates to optimize network utilization and reduce storage costs in video security applications
  • High frame rate encoding – Ultra HD at 60 frames per second for sports and flying camera applications
  • H.265 solutions delivering up to twice the compression efficiency of current H.264 camera SoCs