(Hong Kong, December 17, 2024) - A wholly-owned subsidiary of Intron Technology Holdings Limited ("Intron Technology" or "Group"; stock code: 1760.HK), a leading provider of automotive electronics solutions in China - Shanghai G-Pulse Electronics Technology Company Limited, released a centralized millimeter-wave radar AI perception algorithm solution based on the Horizon Journey®6 intelligent driving platform. It uses advanced AI algorithms to enhance the performance of the sensing system, and further promotes the precise detection and integration of radar and scenarios, bringing a whole new experience to intelligent driving.
The traditional automotive millimeter-wave radar receives electromagnetic wave signals reflected by objects, and inputs them into a computing unit for identification calculation. The output data contains multi-dimensional information such as the range, class, velocity and intensity of the measured object. Traditional signal processing methods are unable to completely filter out the noise in the raw radar data, and some effective information may be lost during processing, resulting in missed detections and false detections when the radar detects objects.
Based on the Horizon Journey®6 E/M intelligent driving platform and the centralized radar perception system, Intron Technology uses data as the driving force to maximize the collection of effective information, and builds a deep learning computing model with neural network algorithms as the core and continuously optimizes it. Intron Technology converts the collected radar ADC data into two-dimensional and three-dimensional spectrum data. Based on the specific structure of the spectrum data and its extensive years of experience in AI research and development, Intron Technology has developed a perceptual model for efficient processing of radar spectrum data and uses convolutional neural networks in the perceptual model. The network extracts effective information from radar spectrum data. Compared with the traditional rule-based signal processing method, Intron Technology uses a data-driven approach to extract information from more primitive radar data, combined with the radar perception model, to effectively reduce the loss of radar information extraction, and make the radar perception result more accurate. At the same time, Intron Technology's self-developed radar perception algorithm adopts a multi-task learning model structure based on a shared backbone network, which can ensure that the model can output object detection, semantic segmentation, occupancy prediction, and object tracking while under the condition of real-time operation of the model.
 (A) Radar two-dimensional spectrum data |
 (b) Radar three-dimensional spectrum data |
| Two-dimensional spectrum data and three-dimensional spectrum data of Intron Technology centralized radar (The color indicates the intensity – the darker the color, the greater the intensity, and the yellow-green area indicates the target area) |
|
Based on its in-depth understanding of Horizon Journey®6 series chips, Intron Technology optimizes the algorithm architecture to maximize the computing potential of the chip. Traditional radars usually directly output radar perception results, and perform late-fusion perception with other sensors such as cameras and LiDARs in the intelligent driving domain controller. The results are limited by the perception capabilities of each modality. In order to fully tap the raw data of the sensor and improve the perception performance, the Intron Technology radar perception algorithm solution uses the Horizon Journey®6 series chip DSP and BPU resources to process the raw data from the millimeter-wave radar, and integrates the optimized FFT (Fourier Transform) and IFFT (Inverse Fourier Transform) and other algorithms into the DSP unit. It also deploys 3D detection, segmentation, occupancy grid and tracking algorithms on the BPU, which improves the perception accuracy while ensuring the real-time performance of data processing. The solution can also support three multi-sensor fusion sensing algorithms of early, middle and late, which can be adapted and integrated with the sensing algorithms of different routes to achieve higher sensing performance.
Schematic diagram of the Horizon Journey ®6E/M radar data processing flow
(DSP is responsible for traditional radar signal processing, BPU is responsible for deep learning perception algorithm processing)
The Intron Technology domain controller solution supports up to seven cameras, four centralized corner radars and one front radar, plus 12 ultrasonic sensors. The domain control solution is designed for high-speed NOA and urban NOA scenarios, and supports the integration and deployment of AI-based millimeter-wave radar sensing algorithms. While improving sensing performance, system costs can be further reduced. In terms of sensor interface, the camera and the centralized millimeter-wave radar adopt a unified interface, which can be flexibly adjusted to further expand the application scenarios. In terms of functional safety, the program integrates a system safety monitoring mechanism that can monitor the operating status of each part in real time and quickly locate and repair faults, thereby improving the safety of automotive operation.
Schematic diagram of the Intron Technology Domain Controller
As the automotive electronic and electrical architecture evolves from a distributed architecture to a domain architecture and then to a central computing architecture, centralisation of automotive computing power has become a general trend. Radar, as the core of the intelligent automotive perception system, is also evolving toward super-sensors in this evolution, which is the direction that the centralized millimeter-wave radar has been derived. It transfers the computing unit to the domain controller, and uses the abundant computing power of the domain controller to provide greater flexibility for multi-sensor data fusion, making the radar lighter and more sensitive, and significantly reducing the hardware cost of the radar system. The Intron Technology centralized millimeter-wave radar algorithm solution will greatly improve the operational efficiency of the perception system, achieve highly robust, high-precision 4D object detection results, and facilitate efficient, intelligent as well as safe driving.
-End-
About Intron Technology Holdings Limited (Stock code: 1760)
Intron Technology Holdings Limited is a leading automotive electronics solutions provider in China focuses on providing solutions targeting critical automotive electronic components applied in New Energy, Automated Driving, Connectivity, Body Control, Safety and Powertrain systems. The Group leverages its engineering and R&D expertise, combined with advanced semiconductor devices, to offer industry-leading productized solutions for customers, thereby fostering the sustainable development of Automotive Industry in China. Intron Technology has been included in the MSCI China Index and the FTSE Global Equity Index Series ("GEIS").
Media Inquiries:
Strategic Financial Relations Limited
| Veron Ng | Tel: 2864 4831 |
| Iris Au Yeung | Tel: 2114 4913 |
| Lilia Yang | Tel: 2864 4833 |
| Email: sprg_intron@sprg.com.hk | |
