On December 20, 2017, after two years of relentless effort, Horizon Robotics, a world-leading AI startup, successfully launched China’s first globally leading embedded AI “Chinese chip”. This innovative solution, under the banner of “Algorithm + Chip + Cloud”, aims to provide a groundbreaking Chinese solution. In the field of AI visual recognition, this type of chip can detect up to 200 visual targets simultaneously in each frame, providing foundational support for smart driving and smart city development in China.
At the launch event, Horizon’s founder & CEO, Yu Kai, along with Chinese Academy of Sciences academician and Tsinghua University computer professor Zhang Bo, and Ding Wenwu, president of the National Integrated Circuit Industry Investment Fund, jointly released the Journey (Journey) 1.0 processor for smart driving and the Sunrise (Sunrise) 1.0 processor for smart cameras. These are AI chips entirely developed by Horizon, utilizing Horizon’s first-generation BPU architecture, boasting world-leading performance: capable of real-time processing 1080p30 video, detecting, tracking, and recognizing up to 200 targets per frame, with a typical power consumption of 1.5W and a frame delay of less than 30ms.
It is reported that the Journey 1.0 processor and the Sunrise 1.0 processor are based on Horizon’s first-generation BPU architecture — the Gaussian architecture. The Journey 1.0 is aimed at smart driving, capable of accurately monitoring and recognizing various targets in real-time, including pedestrians, motor vehicles, non-motor vehicles, lane markings, traffic signs, and traffic lights, while meeting the stringent environmental requirements for automotive applications and the visual perception needs in different environments. The Sunrise 1.0 processor is designed for smart cameras, capable of large-scale local face capturing and recognition, video structuring processing, and can be widely used in various practical application scenarios such as commercial and security fields.
Based on these two “Chinese chips”, Horizon has launched corresponding “brain” solutions for smart driving, smart cities, and smart commerce, and conducted relevant technical demonstrations at the launch event.
Among them, the smart driving solution accurately locates targets in driving scenarios to create an autonomous driving brain for vehicles, achieving advanced driver-assistance system (ADAS) functions such as lane departure warnings and vehicle and pedestrian collision alerts. At the launch event, Horizon set up a simulated traffic scenario, allowing guests and media to experience pedestrian detection, achieving an accuracy rate of up to 100%. According to the product manager on site, the current ADAS based on the Journey 1.0 has an accuracy rate greater than 99% for detecting vehicles, pedestrians, lane markings, and traffic signs on actual roads.
The smart city solution can perform high-performance, low-power face capturing, recognition, and related attribute analysis, as well as video structuring analysis at the front end, and can be widely applied in security and general security fields in stations, schools, businesses, buildings, and checkpoints. During the launch, Horizon demonstrated the real-time face capturing and recognition capabilities of a camera based on the Sunrise 1.0 processor, accurately capturing and recognizing the speaker, Dr. Zhou Feng, Chief Chip Architect of Horizon, from hundreds of people under complex lighting conditions using a mobile camera. An outdoor exhibition area was also set up for real-time large-scale face capturing experience, capable of capturing over a hundred faces simultaneously.
The smart commerce solution focuses on structuring offline commercial operation data centered around people, achieving customer flow analysis, personnel ID management, and goods analysis, helping commercial operation systems gain deeper insights into the essence of business operations, guiding marketing strategies, and enhancing operational efficiency, thereby improving the consumer shopping experience and making business operations simpler and more understandable. At the event, Horizon showcased a practical use case of this solution in a shoe store, where the camera could real-time identify customer identities, preferences, and their movement trajectories within the store.
