Research And Implementation Of Smart Helmet Technology Based On Embedded Linux

With the development of my country’s construction industry,the issue of personal safety on construction plants has become gradually acute.Additionally,the wide application of artificial intelligence chips and deep learning technology makes it possible to apply artificial intelligence technology in embedded systems.According to the urgent needs of safety construction and information management,this paper carries out technical research and functional realization of the intelligent safety helmet based on embedded Linux.Firstly,this essay elaborates the research background of the subject of smart helmets,briefly describes the domestic and overseas research status of smart helmets and investigates the research and development status of embedded artificial intelligence equipment.After that,the demand for smart helmets is analyzed,and the research significance is shown.The integrated design of the smart helmet system is elaborated,including the core processor,the prime control module,the hardware peripheral equipments and the pivotal streaming media technology.Subsequently,according to the demands of the smart helmet,the general planning of the system software and the explicit functions of the software at each level are established.After transplanting the system boot program,cutting the operating system kernel,and making the Jffs2 root file system,the Linux operating system is implemented on the embedded platform.The transplantation of peripheral equipment and the construction of the development environment have entirely constructed the software platform of the smart helmet system.This dissertation is committed to the video system which is the basis for realizing the functions of the smart helmet,including the complete process of video data acquisition,input,processing and encoding.Aiming at the situation that the wireless network channel quality is easy to fluctuate,a data rate adaptive adjustment function is designed,and the Padhye model and adaptive adjustment strategy for network rate estimation are introduced.Combined with RTSP streaming media protocol,the video adaptive adjustment part is constructed,which adapts to the current network transmission quality by changing the video resolution according to different network conditions.Then,in order to realize the function of auxiliary construction site sign recognition,a video-based deep learning target recognition algorithm is studied.Aiming at the less resources of the embedded platform and the feasibility of transplantation in this subject,the requirement for lightweight improvement of Yolov5 s network structure is put forward.The lightweight network structure is introduced to improve the backbone feature extraction network in the original model,and the compression-excitation module in the channel attention mechanism is introduced to improve the network recognition accuracy.On the precondition of reaching the discernible precision,the number of network layers and the structural complexity are greatly simplified to realize the lightweight improvement of the Yolov5 s network model.Finally,build the embedded verification platform of the subject and verify the system effect.The video system part of the adaptive data transfer function is tested.The performance of the construction site sign recognition algorithm is tested.