تشخيص بلادرنگ اشياء در تصاوير ديجيتال با استفاده از يادگيري عميق
چکيده
تشخيص شي يکي از مهمترين و پرچالش ترين شاخه ي فعاليت هاي بينايي ماشين است که حوزه ي وسيعي از زندگي مردم را پوشش مي دهد. تشخيص شي به دسته بندي و مکان يابي اشيائي خاص در تصوير مي پردازد و يکي از کاربردهاي مهم آن، احراز هويت و تشخيص چهره است. از آن جايي که شبکه هاي عصبي عميق در تشخيص شي عملکرد بهتري نسبت به روش هاي سنتي دارند، در اين پايان نامه الگوريتم يولو به عنوان يک شبکه ي عصبي عميق پياده سازي مي شود. اين شبکه به صورت بلادرنگ عمل تشخيص شي را در تصاوير ديجيتال انجام مي دهد و يکي از سريعترين الگوريتم هاي تشخيص شي است. در اين پايان نامه الگوريتم يولو را براي تشخيص چهره توسط مجموعه داده ي WIDER FACE آموزش داديم و دقت آن براي سه دسته داده ي ساده، متوسط و دشوار به ترتيب 50/17، 45/53 و 36/72 درصد محاسبه شد.
نگارنده: ساناز خليلي (khalili.sanaz94@gmail.com)
اساتيد راهنما : آقای دکتر محمدرضا حيدريان (mrh@vru.ac.ir) - آقای دکتر علي شکيبا (ali.shakiba@vru.ac.ir)
اساتيد داور : آقای دکتر مجتبي صباغ جعفري(mojtaba.sabbagh@vru.ac.ir) - آقای دکتر سيد جلال سيد يزدي(seyyedyazdi@vru.ac.ir)
دانشگاه وليعصر(عج) رفسنجان - www.vru.ac.ir
The most important and challenging field of computer vision is object detection. It covers a wide range of people's life. Object detection includes the classification and localization of special objects in pictures. The most important applications are biometric authentication and face detection. The performance of deep neural networks is better than traditional methods in object detection. In this thesis, we implement the YOLO algorithm that is a deep neural network and detects objects in digital images in real-time. YOLO is one of the fastest object detection algorithms. We used the WIDER FACE dataset for training YOLO and calculated the mAP for the easy, medium, and hard datasets as 50.17, 45.53 and 36.72, respectively.
By: Sanaz Khalili(sanaz.khalili@stu.vru.ac.ir)
Supervisors: Dr. Mohammadreza Heydarian(mrh@vru.ac.ir) , Dr. Ali Shakiba(ali.shakiba@vru.ac.ir)
Examiners: Dr. Seyyed Mojtaba Sabbagh(mojtaba.sabbagh@vru.ac.ir) , Dr. Seyyed Jalal Seyyed Yazdi(seyyedyazdi@vru.ac.ir)
Vali-e-Asr University of Rafsanjan - www.vru.ac.ir