The KL630 series chips are built using TSMC’s 28nm advanced technology node and are designed for a wide range of professional and consumer IP cameras. These include security cameras, home cameras, video doorbells, stabilization cameras, 180° and 360° panoramic cameras, and cameras integrated into home appliances and robots. The KL630 series supports the rapid development of high-quality, multi-camera applications. It features up to four-channel video input, independent advanced multi-ISP pipelines (including multi-exposure high dynamic range fusion), and specialized video quality enhancement functions for multiple wide-angle lenses and sensors. The chip also includes on-chip distortion correction, stitching, and blending, which all contribute to real-time high-quality panoramic video processing.

With Kneron’s third-generation neural network core embedded, the chip can deliver real-time object recognition, reducing false detection rates.

Key Features

• Reconfigurable NPU: Kneron is a leader in the field of reconfigurable NPUs. They have pioneered the introduction of dynamic memory access (DMA) to improve memory access efficiency and dynamically support different data precision requirements within the same neural network. This allows the NPU chip to deliver ASIC-level performance without sacrificing programmability for data-intensive algorithms.
• With its innovative architecture and excellent performance, the Kneron team was awarded the IEEE CAS 2021 Darlington Best Paper Award.
• The 4th generation reconfigurable NPU can simultaneously run CNN and Transformer networks, enabling both machine vision and semantic analysis tasks. Its outstanding computing efficiency provides end users with a higher-performance, low-power, and cost-effective solution suitable for AI applications in a wide range of terminal devices.

• Unlike standard AI models that target specific applications, Kneron’s Reconfigurable Artificial Neural Network (RANN) technology offers greater flexibility, allowing it to meet various application needs and adapt to different computing architectures.
  
• The characteristics of Kneron's image recognition software align with those of its hardware products. In addition to offering low latency, high computational power, and strong performance, Kneron’s AI models and algorithms balance performance with model size and complexity.

• The Naineng lightweight image recognition algorithm mainly includes features like facial recognition, body and gesture recognition, and object and scene recognition. These models are already included in the release of the standard SDK developed by the company.
  
• The fifth-generation ISP can capture images and videos with perceptible details, true-to-life tonal range, and reduced noise. Its key features are as follows:
  
• "Color Consistent" technology dynamically compensates to reduce color discontinuities and intensity variations.
• "HDR" technology offers clearer details in high-contrast scenes, presenting richer and more accurate colors to deliver a more lifelike visual experience.
• "Smart Edge Enhancement" technology improves localized image blur (defocus) caused by slight misalignments of the lens and sensor.
• "Multi-lens and sensor ISP" technology compensates for color and brightness discrepancies in video stitched or fused from multiple lens and sensor sources due to production differences.
• The intelligent HEVC encoder significantly reduces unnecessary or unimportant data while maintaining high-quality video without straining network bandwidth. With the latest advancements in dynamic ROI, GOP, FPS, and noise reduction, it intelligently detects regions of interest, particularly in dynamic scenes with varying motion. It applies noise reduction at different intensities, adjusts the group of pictures (GOP) and frame rate accordingly, ensuring optimal bit rate savings without losing key details. Compared to H.264 encoders in video surveillance, the smart HEVC encoder achieves extremely low bit rates, saving up to 60% on network bandwidth and local storage costs.

Fisheye Engine for De-warping, Stitching, and Blending

The support for 180°/360° multi-channel fisheye de-warping introduces an innovative approach, combining high-performance hardware-accelerated fisheye de-warping that allows real-time processing, significantly reducing CPU load. It flexibly supports geometric correction for various fisheye and wide-angle lenses, delivering industry-leading panoramic video quality.

De-warping  

• Supports up to 4 video inputs
• Algorithm optimization based on pixel/block transformation

Stitching and blending  

• Feature-based matching
• Geometric model stitching
• Compensation for color deviations between different sensors

Intelligent Video Analysis

"AI On-Device" is a neural network-based design that analyzes real-time video, classifying trained objects such as people (especially faces), pets, cars, and other objects of interest for detection and recognition. This enables end users to benefit from enhanced privacy protection, data localization, and lower data access latency when responding to tasks run on the device itself.

Seamless customization is integrated with the built-in ISP and de-warping engine, allowing high-precision detection and recognition even when using wide-angle or fisheye lenses.

Audio Enhancement and Intelligence

Creative machine learning techniques are applied to enhance speech and audio signal processing. VATICS audio technology has inspired innovative and elegant solutions to address some of the most challenging and longstanding technical issues in the field, such as noise suppression, acoustic echo cancellation, automatic gain control, and dynamic range control.

Key advantages include:

• Competitive performance with a smaller footprint, reducing computing resources by up to 30% compared to other solutions on the market.
• Lower hardware design constraints.
• A machine learning-based internal engine that significantly reduces the need for manual adjustments.

High-Performance Application Tools: PQ Tool and Fisheye Correction Tool

The PQ Tool offers a fast and comprehensive image adjustment solution.

EZ tuning offers 10 presets aligned with 17 blocks to accommodate ISO variations in the current architecture. You can adjust the values either by typing or by directly dragging the sliders. Please refer to the attached image for details.

The Fisheye Correction Tool allows for real-time adjustments and immediate preview.

The KL630 series chips offer design flexibility, with options for both external and built-in memory, based on PCB design considerations.

Kneron provides a block diagram and key AP functions simultaneously, allowing customers to get started quickly.

Kneron provides a block diagram and key AP functions to help customers get started quickly.

Kneron provides a block diagram and key AP functions, enabling customers to quickly get started.

KL630 Series AI Cameras (with Stabilization Cameras)

In surveillance systems, cameras mounted on tall poles, vehicles, or mobile devices can experience shaking due to wind, motion, or hand movements.

Digital/Electronic Image Stabilization (DIS/EIS) is an effective video stream correction method that reduces video shake and rolling shutter effects. It relies on gyroscope sensors to produce angular velocity signals for EIS, which then corrects the image by cropping and resizing it within the current frame. The MotionProcess library computes a distortion matrix for ImgSigProcess (ISPE) to perform video stabilization and rolling shutter correction, enhancing image quality.

The main principle uses triaxial angular velocity and acceleration. Over a period of time, the gyroscope sensor accumulates the total angular displacement and calculates how much angle adjustment is needed per frame. The process is illustrated in the diagram below:

Please note that horizontal/vertical shifts will not affect the gyroscope sensor. The gyroscope should be kept as far away from the lens as possible.

Kneron’s solution also allows for different configuration adjustments based on the position of the image. As shown in the diagram below, adjustments such as the RATIO and fisheye correction can be applied, considering the image position and adjustments.

Source: Kneron Official Website: Kneron - Artificial Intelligence Everywhere

Application scenario image

Display board photo  

Solution block diagram

Core Technology Advantages:

1. Reconfigurable NPU
2. Image recognition software
3. Fifth-generation ISP for capturing images and videos with perceptible details, true tonal layers, and reduced noise
4. Intelligent HEVC encoder significantly reduces unnecessary or unimportant data while maintaining high-quality video without interrupting network bandwidth
5. Fisheye engine for de-warping, stitching, and blending
6. Intelligent video analysis
7. Audio enhancement and intelligence
8. High-performance application tools: PQ Tool and Fisheye Correction Tool

Solution Specifications:

• KL630 Series AI Cameras
• KLR5S3 Series (Car DVR)
• KLI5S3 Series (Industrial Cam)