Object distance analysis using convolutional neural network (CNN) method based on stereo vision

B. Hilda Nida Alistiqlal, I Wayan Sudiarta, Susi Rahayu

Abstract

Blind or semi-blind people require additional devices or assistants such as a cane or a guide dog to navigate in their daily activities. In robotic or automation technology such as self-driving cars also require knowing objects in front of them accurately. The purpose of this paper is to study the use of stereo cameras as a distance measuring device and to determine the optimal camera configuration and its accuracy. The method used to determine distance is the convolutional neural network (CNN) method based on stereo vision. For training and validation data, we recorded images of a ball in random positions in front of the stereo cameras. This test uses two steps, the first is to recognise the camera object using the CNN method, then the second is to measure the stereovision-based object with an input image of 1350 image pairs. With 160 images used as CNN training data and the remaining 40 images as CNN validation data, the entire dataset is also taken to be used as training data for ball distance. Based on the results of the study, it was found that the model successfully predicted the image with 97% accuracy. The accuracy of the optimal distance measurement results at a distance between cameras of 11 cm is found to be 68.03% within a distance of 1 m, for a distance between cameras of 15 cm is found to be 94.63% within a distance of 2 m, and for a camera distance of 19 cm is found to be 99.61% for a distance of 3 m objects.

Keywords

Convolutional Neural Network (CNN); Measuring Distance; Stereo cameras.

References

Arsada, B. (2017). Aplikasi Sensor Ultrasonik Untuk Deteksi Posisi Jarak Pada Ruang Menggunakan Arduino Uno. Jurnal Teknik Elektro, 6(2), 1–8.

Ramadhan, R. (2012). Sistem Pendeteksi Objek untuk Keamanan Rumah dengan Menggunakan Sensor Infra Red. 1–17.

Sabirin, M. S., Magdalena, R., & Susatio, E. (2020). Prototipe Pendeteksi Jarak Menggunakan Kamera Smartphone Dan Laser ( Prototype Distance Detection Using a Smartphone Camera and Laser With Hsv Color Model ). 7(2), 4127–4133.

Kusumanto, R. D., & Tompunu, A. N. (2011). Pengolahan Citra Digital Untuk Mendeteksi Obyek Menggunakan Pengolahan Model Normalisasi RGB. Semantik, 1(1), 37–72.

Dewi, R. S. (2018). Deep Learning Object Detection Pada Video. Deep Learning Object Detection Pada Video Menggunakan Tensorflow Dan Convolutional Neural Network.

Valueva, M. V., Nagornov, N. N., Lyakhov, P. A., Valuev, G. V., & Chervyakov, N. I. (2020). Application of the residue number system to reduce hardware costs of the convolutional neural network implementation. Mathematics and Computers in Simulation, 177, 232–243. https://doi.org/10.1016/j.matcom.2020.04.031

Adhiem, M. A. N., Nazaruddin, Y. Y., & Zahra, N. (2021). Pengembangan Trem Otonom Tanpa Rel. Jurnal Otomasi Kontrol Dan Instrumentasi, 13(2), 125–133. https://doi.org/10.5614/joki.2021.13.2.8

DOI: https://doi.org/10.20961/jphystheor-appl.v8i2.86069

Refbacks

  • There are currently no refbacks.