Interactive Portrait Acquisition System Based On Continuous Pose Feedback

Convenient,quick,and high-quality portrait acquisition has evolved into one of the basic necessities of the public with the growth of the digital society and the pervasive usage of personal biometric identity in daily life.The traditional portrait acquisition method has the drawbacks of low efficiency and convenience while high quality;the current self-help portrait acquisition device and cell phone self-help photography,although easy to operate and high efficiency,are generally of low image quality and difficult to present as a good personal image.This research develops a dynamic system for acquiring portraits based on ongoing pose input in light of the quick advancement of intelligent perception technologies.The system may conduct customized processing and achieve intelligent engagement during the portrait acquisition process through ongoing interactive posture suggestion.Through ongoing interactive pose prompting,the system achieves intelligent interaction during the acquisition of portraits and can process and optimize portrait images according to user preferences.The following are the paper’s primary contents:First,to enhance the applicability of the portrait acquisition system and to ensure that the system can accurately extract the photographed person as the interaction target in increasingly complex scenes.In this paper,we add saliency(SANet)module to YOLOv5 s target detection algorithm to extract image saliency information and enhance the feature representation capability of the network for the target;add convolutional attention(CBAM)module to make the network focus on important information;and introduce EIoU loss function instead of CIoU loss function to improve the detection capability of the model and solve the problem of low robustness of the human-computer interaction system in complex scenes.The problem of low robustness of target object detection in complex scenes is solved.Secondly,in order to realize the prompt correction of irregular movements of the human-computer acquisition system and ensure that the system can continuously and stably identify the real-time action posture of the interaction target.In this paper,we use the Alphapose algorithm to extract the skeletal points of relevant actions to organize into skeleton sequence data,and construct a data set containing seven kinds of irregular actions,and add the channel attention(SENet)module on the basis of spatio-temporal graph convolutional neural network(ST-GCN)to improve the model’s ability to capture important features.Again,to further enhance the interaction details and improve the quality of portrait images,and to ensure that the system can meet the user’s demand for portrait images.In this paper,the Dlib tool library is used to extract the key points of faces and implement the key point-based facial pose detection function;meanwhile,the OpenCV library is used to implement the functions of portrait cropping,portrait optimization and quality detection.Finally,the software and hardware design of the interactive portrait acquisition system based on continuous pose feedback and its core functions are completed.The system selects hardware devices such as camera,exposure lamp and PC according to the requirements of pose detection and portrait processing functions required for portrait acquisition,and completes the functions of portrait verification and automatic crop of passport portraits required in the actual use process.Experimental results show that the system has stronger applicability,interactivity and practicality than the current existing portrait acquisition methods.