Based on the FPGA platform, self-propelled technology continues to mass-produce intelligent driving solutions such as advanced driver assistance systems (ADAS) and intelligent cockpit systems (IMS) with high reliability and high cost performance.
Shenzhen Self-Propelled Technology Co., Ltd. is a world-leading provider of intelligent driving solutions and equipment, and is also a Pioneer Enterprise in China developing computer vision and sensor fusion technology based on FPGA hardware platform, with deep talent and technical reserves in the field of FPGA-enabled image processing technology and deep learning technology, as well as a perfect product development tool chain. At the same time, the company is also one of the national standard drafting units for "Driver Attention Monitoring System Performance Requirements and Test Methods" (known in the industry as DMS) in China, a software partner of Bosch in China, and one of toyota Tsusho and Xilinx certification partners.
The core technology of self-technology
Project overview
The development of artificial intelligence, 5G and internet of vehicles is redefining the safety and human-computer interaction experience of future cars, and the iteration and deployment of various intelligent driving technologies corresponding to them are also accelerating.
"Intelligent driving, ADAS first". As the two important areas of intelligent driving that have been launched first, ADAS and IMS are promoting the landing process of intelligent driving from the two directions of "driving automation" and "cockpit intelligence".
Self-propelled technology is one of the earliest enterprises in China to contact FPGAs, and is also one of the earliest certified partners of Xilinx in the automotive market in China. With The Helpers Automotive Specification Zynq-7000 SoC Series Adaptive Computing Platform, Self-Propelled Technology has continued to mass-produce a series of intelligent driving solutions for visual perception inside and outside the vehicle such as ADAS and IMS. The Dynamic Function Conversion (DFX) function of the Xilinx Zynq platform provides higher performance, lower latency, and the flexibility of multiple ADAS functions to load and dynamically convert on unified computing resources in real time.
Industry challenges
The Surround View System (AVM) is one of the most permeable ADAS features in recent years, also known as the 360-degree Surround View System. The system senses the environmental data outside the vehicle through four camera sensors installed on the front, rear, left and right of the vehicle, helping the driver understand the environment around the vehicle and providing timely reminders before dangers occur. Compared with a simple front-view or rear-view system, AVM needs to process data from 360° of the body in real time, while helping drivers make quick decisions through image stitching, video processing and other technologies.
Traditional AVM systems are usually implemented based on ARM or GPUs, for example, NXP's I.MX6 family of chips. These chip platforms will face various limitations when applied to AVM systems, such as arm low efficiency, high GPU power consumption, slow startup time and high cost, making it impossible to meet the ultra-large-scale deployment needs of intelligent driving in depots in terms of energy efficiency and cost.
At the same time, in the field of intelligent cockpit, with the increasing development of end consumers' demand for diversified interactive experience of intelligent cockpit, in addition to traditional voice and touch interaction, intelligent interaction based on visual sensors and multi-sensor fusion intelligent interaction are also increasingly loved by consumers. In the face of diversified and continuously changing experience needs, intelligent needs, sensor fusion requirements, coupled with timeless cost requirements, in the limited cockpit environment, the existing ASIC, ASSP, DSP and other fixed or standard processor platforms are difficult to meet all of the above needs in terms of size, performance, cost and other aspects.
With rich experience in FPGA design, its own technology is the first to realize that FPGAs have the characteristics of flexible deployment, fast startup speed, strong stability, rich logic resources, and adaptive provisioning, which can well meet the various functions required by AVM and IMS systems. They hope to combine their own world-leading AI technology and excellent FPGA development capabilities to create a new generation of AVM and IMS systems for the industry.
solution
Self-Propelled Technology selected Xilinx Automotive Specification Grade (XA) Zynq-7000 SoC Series Adaptive Computing Platform to create a new generation of AVM and IMS systems.
Dr. Guan Yanfeng, CEO of Self-Propelled Technology, said: "As a heterogeneous multiprocessor hardware processor platform based on FPGAs, Zynq SoCs integrate programmable logic (PL) and on-chip processing systems (PS) into one, and not only have the functions of general processors such as ASICs, ASSPs and DSPs, but also have more functions that other processors do not have. These capabilities, combined with our leading AI deep learning technologies and computer vision and sensor fusion technologies, bring valuable differentiation and leadership to our next-generation AVM and IMS systems. ”
In response to customer needs, Based on Xilinx's mature vehicle specification-grade Zynq-7Z010 platform, self-technology has developed an efficient AVM system, which realizes fast start-up speed, low splicing distortion and high robustness, and at the same time has the functions of dynamic follow-up of the viewing angle and dynamic loading of front and rear auxiliary lines, which injects a fresh impetus into the red sea market of the traditional AVM system. At the same time, with the help of its own leading AI algorithm capabilities, self-technology has developed a cost-effective and rich variety of IMS based on the Zynq-7Z010 and 7Z020 platforms, with a wide range of functions, including fatigue detection, attention detection, identity recognition, status recognition, behavior recognition, gesture recognition, emotion recognition, occupant status, rear recognition, etc.
The new generation of AVM and IMS systems in-house technology achieves:
Higher performance. A single Zynq chip can realize a complete ADAS solution design from environmental characterization to sensor fusion and feature extraction, so that the self-technology solution avoids the bandwidth bottlenecks and security risks of chip-to-chip transmission of all other cross-chip platform solutions, so that the solution performance is higher, more stable and reliable.
Lower latency. Real-time response capabilities with low latency are critical to driving safety. The unique FPGA parallel processing benefits of Xilinx SoCs allow for flexible access to a wide range of sensors and parallel processing of data from the entire front-end sensor in nanosecond or microseconds, with lower system latency than the GPU or CPU.
Lower Total System Cost of Ownership (TCO): Self-propelled technology makes full use of Zynq's rich hardware resources such as BRAM, lookup tables, registers, and DSPs to successfully deploy computationally complex deep learning networks on low-cost Zynq chips, successfully developing cost-effective, versatile, and highly competitive IMS and ADAS systems.
More flexibility. With the unique reconfigurable capabilities of FPGAs, as well as the deep learning modular design and recombination technology of auto technology, the new solution of auto technology can support users to customize new functions according to their needs, or easily expand sensors. At the same time, it also laid a solid foundation for the rapid optimization and deployment of new deep neural networks by its own technology, as well as the subsequent expansion of various ADAS system functions.
Smarter and more secure: Zynq SoC's DFX capabilities enable a new solution from auto technology to enable real-time loading and dynamic transformation of multiple ADAS functions on a unified computing resource. For example, other driver assistance functions such as lane departure warning system (LDW) and blind spot monitoring system (BSD) are implemented on the new AVM system to make driving smarter and safer.
Layout of intelligent products of self-propelled technology vehicles
Design results
Based on the FPGA hardware platform, self-technology continues to develop cost-effective intelligent driving solutions such as ADAS and IMS. Compared with the traditional I.MX6D solution, the new AVM system of self-propelled technology reduces the cost by more than 50% compared with the traditional chip (I.MX6D solution) on the main chip alone, and realizes the current level of mainstream intelligent automotive electronic product solutions in terms of function and performance, which can meet the needs of the smart automotive electronic product market at home and abroad.
At present, all kinds of intelligent driving electronic products and solutions developed by self-technology based on the FPGA platform have successfully achieved stable mass production in the front-loading market of commercial vehicles and passenger cars (China's largest bus brand, as well as well-known joint venture automobile brand models with annual sales of 2 million, etc.), and the cumulative shipment has reached tens of thousands of units.
Bus DMS product schematic
Automotive AVM product schematic
(7870912)