基于机器学习的Cortex-M监控视频车型识别

为了解决高速公路环境下监控视频图像车型识别需要将海量视频数据上传计算机服务器中心,对所有的视频流图像进行结构化处理和车型识别,造成服务器中心数据压力大、计算任务重,对服务器性能要求高的问题。对此,提出一种基于机器学习的Cortex-M监控视频车型识别的方法。首先,将训练机训练好的六种车型的权值矩阵文件移植到前端Cortex-M核系列开发板上,采用开发板内嵌的CMSIS-NN网络函数库搭建与训练机相同的网络模型结构;同时采用开发板内嵌的CMSIS-DSP库加快图像处理速度,并对选择处理监控视频图像实现车型识别;实验结果表明,该方法平均识别率达到94.6%以上,与采用计算机进行识别相同,可见该方法能够缓解大量视频上传给服务器中心造成的压力,为高速公路环境下监控视频图像车型识别研究提供了一种可选择的方案。

Research on Cortex-M surveillance video vehicle type identification based on Machine Learning

In order to solve the problem of surveillance video image vehicle type recognition under the expressway environment, it is necessary to upload massive video data to the computer server center and structured processing vehicle type recognition for all video stream images, which causes great pressure on the data in the server center, heavy computing tasks and high requirements on the server performance. Based on this, a method of Cortex-M surveillance video vehicle type recognition based on machine learning was used to investigate. Firstly, the weight matrix files of the six models trained by the training machine were transplanted to the front-end Cortex-M core series development board and the CMSIS-NN network function library embedded in the development board was used to build the same network model structure as the training machine; At the same time, the CMSIS-DSP library embedded in the development board was used to speed up the image processing speed and the vehicle type recognition was realized for the surveillance video image of the selected processing. The experimental results show that the average recognition rate of the method is above 94.6%, which is the same as the recognition by computer. It is concluded that this method relieves the pressure caused by a large number of video uploads to the server center and provide an alternative solution for surveillance video image vehicle type recognition in the highway environment.