Last summer, we reported about DMP Inc. and Basemark's announcement of smart mirrors for vehicles using AI. However, that was just the tip of the iceberg for DMP’s AI efforts. In May 2019, Yamaha Motor ($16 billion revenue) took a 10% stake in DMP Inc. In coming years, Yamaha Motors will ship 50 million motorcycles, boat motors, golf carts, etc., with DMP's AI solutions.
Then in July 2020, DMP introduced its ZIA ISP (Image Signal Processor) IP core. ZIA ISP, says DMP, contributes to higher image quality and higher performance of AI camera devices such as robotic vehicles, drones, and security cameras. It also contributes to highly accurate visual SLAM using a lower cost camera than LiDAR and offers higher performance and cost reduction of customers’ products, says the company.
ZIA ISP is an IP core that performs image processing on the RAW data output from an image sensor. DMP says it accelerates the camera signal processing pipeline in a small size, which performs correction processing for the optical system such as lenses and scratch correction caused by variations in the image sensor for each pixel unit. In addition, by supporting a Red Clear Clear Blue (RCCB) automotive sensor, the dynamic range expansion by C pixels and HDR makes it easier to detect the contours of objects and people, especially in high and low brightness subjects. Since the user interface supports the AMBA AXI interface, it can be easily embedded in ASIC or FPGA. ZIA ISP can be used as an independent IP core, but by combining it with DMP's AI processor ZIA DV720, says the company, a highly accurate and highly efficient image recognition solution can be realized.
The main features and functions of ZIA ISP are:
- Defective pixel correction
- Format conversion
- Gamma correction
- Auto White Balance/Auto gain/Auto exposure
- Frame buffer write
- RCCB sensor compatible
The ZIA ISP uses the MIPI CSI-2 Interface and can handle resolutions to UHD 3840 × 2160@60fps with RAW Bayer/RGB/RCB formats. DMP began shipping the IP in July of 2020. With the major applications being in robotic vehicles, safe driving support systems, drones, security cameras, etc.
To wrap up the package so to speak, in January 2021, DMP introduced ZIA SLAM that simultaneously estimates self-location and creates an environment map. It also supports ROS2, the latest version of Robot Operation System (ROS), a software platform for robot applications, and popularization of robots for various purposes such as AGVs, UGVs, cleaning robots, and drones.
According to a report by Market Insights, the global market size of SLAM robots will grow from $248.5 million in 2019 to $472.6 million by 2025, including automatic guided vehicles, cleaning robots, and AGVs .
And DMP’s ZIA SLAM supports the Nvidia Jetson AGX Xavier as a reference platform. DMP joined the Nvidia Partner Network (NPN) in November 2020.
ZIA SLAM is expected to be applied to a wide range of applications such as automatic guided vehicles, UGVs (Unmanned Ground Vehicles), planetary explorers, submarines, drones, nursing robots, security robots, and service robots.
In early February 2021, DMP added another feature to its robotic vehicle AI-based AV portfolio, ZIA Safe.
JVC Kenwood builds a communication drive recorder for telematics services (a BlackBox for cars), which uses the AI ZIA Safe for functions such as driving support and driver monitoring.
DMP comments that, in recent years, the realization of a safer mobility society has become an urgent issue as long working hours and overworked driving due to the dangerous driving of elderly people and the shortage of drivers in the trucking business have become social problems. To solve this problem, a safe driving support system such as real-time detection of dangerous driving events using the outside camera of the drive recorder and driver monitoring with the in-vehicle camera, and the drive recorder is expected to prevent conventional accidents. It has come to play a role not only as a record but also as a defense measure against risks that occur while driving.
DMP's AI platform ZIA Safe for the safe driving field adopted by JVC Kenwood is a collection of independent modules for object recognition, situation recognition, posture estimation, and other functions necessary for realizing a safe driving support system. The driving support function includes front vehicle collision, start delay, lane departure and signal ignore detections, and the driver monitoring function such as doze detection. It also supports sideways driving and smartphone operation detection while driving. By adopting AI and deep learning technology, says DMP, the platform has achieved high expandability, such as the ability to add detection objects later by learning additional image data of people, traffic signs, and various vehicles.
The company has also applied its technology to AI-based sports coaching. Working with Paris-based GrAI Matter Labs (GML), the companies showcased a novel virtual AI Sports Coaching Application at the Docomo Open House 2021 which was held online from February 4 to February 7.
DMP and GML are developing event-based edge processing solutions for ultra-fast response applications at low power. At the Docomo Open House, the companies showed a virtual AI sports coach that helps golf enthusiasts to improve their swing using Pose estimation.
Pose estimation analysis of humans can track every small movement and do a biomechanical study in realtime. Leveraging GML's GrAI VIP Vision Inference Processor and DMP's edge AI processing algorithms, the companies say a new class of virtual assistants becomes possible that react in realtime and enable pose-based applications in sports analysis, assisted living, industrial automation, surveillance, and more.
What do we think?
DMP is probably best known for its low-power, high-performance 3D Graphics IP with its unique hardware-accelerated pixel and geometry shaders used in the extremely popular Nintendo 3DS. The company has always been clever and pushed the envelope of the technology areas it concentrates on. It has been working in the video ISP segment for several years and combined with its GPU IP. It offers a robust total pipeline IP suite backed up with 20 years of experience. And soon you may be enjoying a ride in a Yamaha AV powered by DMP’s AI, ISP, GPU, and software—AVs bring together all of the company’s tech in one place.
Ironically, DMP has a development center in Ho Chi Minh. It doesn’t have anything to do with Yamaha, who happen to have office near by.