| With the maturity of big data technology and the vigorous development of artificial intelligence technology,intelligent transportation has gradually become an important place for related technologies.Through intelligent transportation technology,the road network can achieve higher operation efficiency,which not only saves people’s time,but also reduces the consumption of resources.In the process of urban development,it is inevitable to encounter a variety of complex problems.For example,urban parking has become an unavoidable problem in the process of urbanization in fast developing cities.At the entrance of the parking lot,a license plate recognition machine is set up to scan and identify the license plate numbers of the vehicles in and out of the parking lot,and connect with the relevant systems and record the vehicle information.License plate recognition can not only strengthen the management of vehicles in and out,but also optimize the distribution of parking spaces and facilitate vehicle owners to query and find parking spaces.Due to the influence of natural environment and monitoring equipment.and other factors,the quality level of collected license plate photos is not uniform.Therefore,it is necessary to accurately identify label information for different quality label photos.This paper focuses on this,aiming at the shortcomings of the existing license plate detection and recognition technology,the problem of license plate photo detection and recognition in the monitoring scene is deeply studied and implemented.This paper studies and implements an automatic license plate recognition system,which consists of license plate location module,character segmentation module and convolution neural network recognition module.This paper mainly studies the use of tensorflow development platform,the construction of CNN convolution neural network,through the use of the collected license plate images,training neural network has good license plate recognition image ability.This paper also introduces the hardware implementation of the license plate recognition embedded system,specifically studies the design and principle of the peripheral circuit of the vehicle recognition embedded system based on stm32mp157 embedded chip,and how to deploy the trained CNN neural network to the embedded platform by using stm32cube-ai software to carry out real-time license plate recognition.Finally,the recognition results are obtained through experiments,and the results are achieved The expected effect is analyzed and the corresponding license plate number is analyzed.The main work and innovation of this paper are summarized as follows:1.The development status of license plate automatic recognition technology is comprehensively and deeply investigated and summarized,which is divided into four stages:license plate location,character segmentation,character recognition and segmentation free license plate recognition.The main methods in each stage and their advantages and disadvantages are clarified.2.A license plate location scheme based on CNN convolution neural network is proposed.3.A simple embedded license plate automatic recognition system is designed and implemented,and the system is tested with its own data set.The system achieves the expected performance in recognition speed and recognition accuracy.Experimental results show that compared with other single function methods,convolution neural network method has higher accuracy and lower miss rate.At the same time,the accuracy of image recognition and license plate location has been greatly improved.Compared with the existing vehicle identification system,the design scheme proposed in this paper not only has higher accuracy and recognition rate,but also greatly reduces the hardware cost of the overall scheme. |
