چکیده
مقدمه
آثار مرتبط
روش های ارائه شده
نتایج تجربی
نتیجه گیری
منابع
Abstract
Introduction
Related works
Proposed methods
Experimental results
Conclusions
References
چکیده
توسعه دوربینهای وب و تلفنهای هوشمند به بلوغ رسیده است و برنامههای بیشتر و بیشتری مرتبط با تشخیص چهره بر روی سیستمهای تعبیهشده پیادهسازی میشوند. تقاضا برای تشخیص چهره بلادرنگ در سیستم های تعبیه شده نیز در حال افزایش است. به منظور بهبود دقت تشخیص چهره، اکثر سیستمهای تشخیص چهره مدرن از چندین مدل شبکه عصبی عمیق برای تشخیص تشکیل شدهاند. با این حال، در یک سیستم تعبیه شده، ادغام این مدلهای شبکه عصبی پیچیده و اجرای همزمان برای دستیابی به هدف شناسایی بلادرنگ چهره انسان و هویت آنها آسان نیست. با توجه به این موضوع، این مطالعه یک مکانیسم تجزیه و تحلیل فریم جدید، مکانیزم پرش فریم های پیوسته (CFSM) را پیشنهاد می کند، که می تواند فریم را در زمان واقعی تجزیه و تحلیل کند تا تعیین کند که آیا تشخیص چهره در قاب فعلی ضروری است یا خیر. از طریق تجزیه و تحلیل CFSM، فریم هایی که برای چهره نیازی به شناسایی مجدد ندارند حذف می شوند. به این ترتیب حجم کار سیستم تشخیص چهره برای دستیابی به هدف تشخیص چهره در زمان واقعی در سیستم تعبیه شده بسیار کاهش می یابد. نتایج تجربی نشان میدهد که مکانیسم CFSM پیشنهادی میتواند سرعت تشخیص چهره را در ویدئو روی سیستم تعبیهشده تا حد زیادی افزایش دهد و به هدف تشخیص چهره در زمان واقعی دست یابد.
توجه! این متن ترجمه ماشینی بوده و توسط مترجمین ای ترجمه، ترجمه نشده است.
Abstract
The development of web cameras and smart phones is mature, and more and more facial recognition-related applications are implemented on embedded systems. The demand for real-time face recognition on embedded systems is also increasing. In order to improve the accuracy of face recognition, most of the modern face recognition systems consist of multiple deep neural network models for recognition. However, in an embedded system, integrating these complex neural network models and execute simultaneously is not easy to achieve the goal of real-time recognition of human faces and their identities. In view of this, this study proposes a new frame analysis mechanism, continuous frames skipping mechanism (CFSM), which can analyze the frame in real time to determine whether it is necessary to perform face recognition on the current frame. Through the analysis of CFSM, the frames that do not need to be re-recognized for face are omitted. In this way, the workload of the face recognition system will be greatly reduced to achieve the goal of real-time face recognition in the embedded system. The experimental results show that the proposed CFSM mechanism can greatly increase the speed of face recognition in the video on the embedded system, achieving the goal of real-time face recognition.
Introduction
The popularity of cameras and smartphones make more and more applications of real-time face recognition on mobile devices, such as using human faces as biometric identifcation [14, 21], or using human faces for access control systems. The corresponding management applications are becoming more and more popular. Previously, face recognition was mainly through image recognition related mechanisms [3, 4, 11, 20, 29]. At present, due to the rapid development of deep learning and neural networks, more and more face recognition mechanisms are implemented based on neural network models. Since the face recognition system usually requires multi-stage analysis, and diferent analysis stages need to use different neural network models. However, the mobile devices and embedded systems cannot provide the enough computation capabilities to process these complex neural network models at the same time, and then achieve the goal of real-time recognition of human faces.
Conclusions
In view of the increasing demand for real-time face recognition on embedded systems, this research is based on a three-stage neural network face recognition system and designed a new frame analysis mechanism, continuous frames skipping mechanism (CFSM), to analyze the video frame in real time. The CFSM system consists of three stages: the abrupt scene change detector (ASCD) to detect abrupt cut and gradual scene; the face recognition interval adjuster (FRIA) to skip a large number of unnecessary face detection frame; and the dark frame detector (DFD) to deal with the dark scenes. The proposed CFSM mechanism can decide whether to perform face recognition and efectively utilize the limited computing resources of the embedded system. The experimental results show that the CFSM mechanism proposed in this study can save up to 90% of the time under the NVIDIA Jetson TX2 system compared to the basic face recognition system.