Alat Bantu Pendeteksi Objek Untuk Tuna Netra Berbasis Ai Mobilenet Pada Raspberry Pi 3B

Abstract

- Over time, technology for disability aids is also growing rapidly around the world. One of the technologies that has a function like the eye is the camera. While in other technologies, now has been created a small PC / computer that can be used as a microcontroller, called a Mini PC modeled Raspberry Pi, but if only using these two tools simple object detection system becomes less complete because the system can only detect objects without knowing the distance, therefore, in order for the distance from an object to be read by the system, the HC SR04 sensor is used which is compatible with raspberry pi, to make this system also needed a data image processing system so that the system can detect objects, in this final task using a pre-trained model of mobilenet, mobilnet is one of the convolutional neural network (CNN) architectures that can be used to overcome the need for excessive computing resources. In this final task the system can detect as many as 80 objects, but in the system test used 5 objects with 50 types of variants, namely bottle objects, glass objects, book objects, people objects, and mobile phone objects, testing was conducted as much as 3 times from each variant of the object with 3 different distances as a determining factor of accuracy, namely distance 100cm, distance 150cm, and distance 200cm, From the results of object detection tests that have been done obtained the average percentage of object detection by 67%, the most accurate objects that can be read by the system at all distances are people's objects, while for other objects only read accurately at a distance of 100cm only.

Pengenalan Objek Menggunakan YOLO pada Alat Bantu Tunanetra Berbasis Raspberry Pi

“…Pada penelitian sebelumnya telah dilakukan menggunakan metode Single Shot MultiBox Detector (SSD) yang dilengkapi dengan buzzer dan vibrator [3] dan algoritma MobileNet untuk menghasilkan sistem yang mampu mendeteksi objek buku, gelas, botol, telepon seluler, dan orang dengan rata-rata keberhasilan pengujian sebesar 40% [6]. Penelitian lainnya terkait algoritma YOLO telah dilakukan dengan menerapkan algoritma tersebut sebagai penentu pindah tanaman cabai melalui notifikasi bot telegram [4], sebagai pendeteksi adanya konten pornografi pada video [5], sebagai algoritma untuk mengenali objek pada citra makanan cepat saji dengan mAP sebesar 100% dan mampu mengenali objek dengan akurasi 63% hingga 100% [7].…”