A Prototype of Visual Navigation Self-driving Car Architecture Based on EAIDK310 and STM32

Abstract

To further understand the AI visual navigation technology, this paper design an AI car by combining STM32 MCU and EAIDK310 developed by OPENAILAB. For the hardware architecture of the intelligent car system, the EAIDK310 development kit equipped with RK3228 is used as the core processor responsible for real-time data acquisition, visual recognition, navigation control, and the STM32F405RGT6 MCU acts as the auxiliary processor to finish IMU data acquisition and instruction execution output. Meanwhile, the dual-camera scheme is adopted to accomplish the automatic driving functioned with road segmentation and traffic sign identification; one camera is used to collect real-time road information at a near distance, and the other one plays the role of capturing environmental information at a wide-angle. For the software system, an efficient, economical, and low-coupling framework is constructed by Redis and Tengine edge inference computing which acting as the critical of data interactive transmission and visual recognition processing, respectively, with this framework, the improvement of real-time and calculating the speed of the system has been realized. Finally, the operation stability of hardware and the software real-time are further proved by experiment measurements, indicating that the proposed method is a potential candidate for a low-cost AI visual navigation system.