Research And Application Of Face Recognition Technology Based On Embedded Systems

This article is based on the project "Safety Violation Intelligent Detection,Identification and Alarm System".In the process of construction management,safety management has always been the top priority.In order to strengthen the safety management of the construction site,before the start of the daily work,the workers need to line up for pre-shift safety education,which is called "station shift" in the construction site management,so as to improve the safety awareness of the workers.For those entering the construction site,it is necessary to strictly restrict and supervise,prohibit the entry of outsiders,and reduce potential safety hazards.Nowadays,face recognition has been widely used.Applying face recognition to construction site management to solve the shortcomings of traditional manual supervision has strong practical application value.Due to project cost constraints,low-cost embedded development boards will be used,with limited memory and computing resources.In this paper,under the limited computing resources of this embedded device,combined with face recognition technology,automatic station status detection is implemented and a face recognition access control system at the entrance of the construction site is designed.The main research work of this article is as follows:(1)In response to the demand for supervision station shifts in construction site management,this paper proposes a face detection algorithm based on the common characteristics of skin color and safety helmet color based on the characteristic that construction personnel wear safety helmets in the actual application environment,and detects camera images in real time In the face,and on this basis,the station shift state detection algorithm is proposed to automatically monitor whether the daily pre-shift safety education is completed.(2)In order to restrict the entry of foreigners to the construction site,a face recognition access control system applied at the entrance of the construction site was designed.Considering that the application scene is outdoor and the lighting conditions are complex and changeable,this paper proposes an image preprocessing algorithm that adjusts the luminance V channel component in the HSV color space,which effectively reduces the face changes caused by workers changing the lighting after wearing a helmet The effect of shadows caused by the occlusion of the brim improves the accuracy of face recognition.Then this paper proposes an improved face recognition algorithm based on FaceNet.The original FaceNet model is a model applied to facilities with strong computing power such as computers and servers.Direct transplantation to the embedded end with limited computing power cannot achieve ideals.Effect.In this paper,the basic network in the FaceNet model is replaced with a SqueezeNet lightweight neural network,and the loss function is improved.The Softmax Loss and Center Loss comprehensive loss function is used for training.Experiments show that the improved FaceNet face recognition model maintains high accuracy Under the premise of the rate,the model parameter quantity and model size are reduced,so that it can achieve real-time face recognition on embedded devices with limited computing and memory resources.(3)A hardware and software environment is built on the embedded development board.According to the above research content,the overall system is designed and implemented,and a simple human-computer interaction interface is also designed.The system is tested,and the experimental results show that the system is stable and reliable,has good real-time performance,and has a good accuracy,which meets the basic requirements in practical application on the construction site.