Design And Implementation Of Embedded Traffic Sign Recognition System Based On Heterogeneous Computing Architecture

In recent years,with the advent of the artificial intelligence era,in order to facilitate people's travel needs,automotive auto-driving technology has developed rapidly.These vehicles are small in size and have limited energy,There are usually two types of problems in their traffic sign recognition systems.When use a general-purpose GPU or GPU computing architecture which has high power consumption.Ability of the endurance of vehicle is affected,when using low power consumption CPU for embedded system,the processing ability is poor,can not meet the requirements of the system to pr^ocess image data in real time.In order to solve the above problems,this paper designs and implements a new traffic sign recognition system based on the heterogeneous computing architecture of embedded system with CPU+FPGA.The system is composed of two segments:a traffic sign segmentation subsystem and a traffic sign recognition subsystem.The traffic sign segmentation subsystem uses the color threshold to segment the traffic sign from the overall image.The traffic sign recognition subsystem uses the tenet-5 convolutional neural network to identify the segmented traffic sign image.The traffic sign segmentation subsystem uses FPGA to accelerate the color gamut space conversion in the color threshold segmentation algorithm,as well as the parallelization and pipeline design of the key points of the algorithm such as blur,expansion and corrosion.The traffic sign recognition subsystem uses FPGA to parallelize the calculation between the convolutional neural network and the convolution kernel.The convolutional layer and the pooled layer are designed as a pipeline structure to accelerate the system.Finally,the system achieves the goal of real-time identification of domestic red and blue background traffic signs from 30fps video stream with a total power consumption of less than 6w,including direction indication,prohibition,speed limit and other 25 types of traffic signs,recognition accuracy.Up to 92%.Compared to the core i5 CPU,and the power consumption is only 17%of the core i5 CPU.