<![CDATA[ABSTRAKVerifikasi kekerabatan berbasis citra wajah merupakan salah satu penerapan sistem Artificial Intelligence yang berguna dalam kehidupan, misalnya untuk penyelidikan kriminal, analisis silsilah, dan lainnya. Perancangan sistem pengenalan wajah pada verifikasi kekerabatan dapat dilakukan menggunakan salah satu algoritma deep learning yaitu metode Convolutional Neural Network. Penelitian ini dilakukan dengan tujuan untuk mengetahui kinerja dari 3D ResNet-18 pada verifikasi kekerabatan berdasarkan sistem pengenalan wajah dan mengetahui kinerja 3D ResNet-18 saat menggunakan embedded system secara real time. Hasil penelitian kinerja ResNet-18 tanpa embedded system memperoleh nilai akurasi training sebesar 0,9771 menggunakan optimizer RMSprop dengan epoch 30 dan batch size 25. Pada pengujian kinerja real time ResNet-18, optimizer SGD berhasil pada ukuran batch size 10, 15, dan 25. Namun untuk pengujian pada perangkat Jetson Nano, optimizer RMSprop gagal akibat ukuran model yang terlalu besar.Kata kunci: embedded sistem, CNN, 3D Resnet18, RMSprop, kekerabatan ABSTRACTFace-based kinship verification is one of the applications of artificial intelligence systems that are useful in various aspects of life, such as criminal investigations, pedigree analysis, and more. The design of a face recognition system for kinship verification can be done using one of the deep learning algorithms, namely the convolutional neural network method. This research was conducted with the aim of determining the performance of 3D ResNet-18 in kinship verification based on face recognition systems and assessing the performance of 3D ResNet-18 when using an embedded system in real time. The results of the ResNet-18 performance research without an embedded system obtained a training accuracy of 0.9771 using the RMSprop optimizer with 30 epochs and a batch size of 25. In real-time performance testing of ResNet-18, the SGD optimizer succeeded with batch sizes of 10, 15, and 25. However, during testing on the Jetson Nano device, the RMSprop optimizer failed due to the size of the model being too large.Keywords: Embedded System, CNN, 3D Resnet-18, RMSprop, Kinship]]>
