Machine learning-based vision system for bomb disposal robot

• Main Author: Galvez, Reagan L.
• Format: text
• Language: English
• Published: Animo Repository 2020
• Subjects: Robots in search and rescue operations; Robotics—Military applications; Explosives—Detection; Electrical and Computer Engineering; Systems and Communications
• Online Access: https://animorepository.dlsu.edu.ph/etd_doctoral/1417; https://animorepository.dlsu.edu.ph/context/etd_doctoral/article/2466/viewcontent/Galvez_Reagan_11788801_Machine_Learning_Based_Vision_System_for_Bomb_Disposal_Robot_1_Redacted.pdf
• Institution: De La Salle University

Bomb disposal or explosive ordnance disposal (EOD) robots are useful in military applications like safe disposal of explosives and search and rescue operations. However, many of these robots cannot identify threat objects using its onboard vision system. As a solution, this study aims to design and develop a vision system for bomb disposal robot using machine learning and image processing algorithms. This vision system can detect and analyze threat objects to help EOD operators in bomb disposal missions. A threat object detector framework was developed composed of two separate modules, such as baggage detection module and improvised explosive device (IED) detection and analysis module. The baggage detection module was capable of wirelessly detecting three classes of baggage, such as a backpack, handbag, and suitcase. The experiments showed that baggage detection achieved 22.82% mean average precision (mAP) using Single Shot Detector (SSD) in the baggage dataset, which was part of the Microsoft Common Objects in Context (COCO) dataset. In addition, the baggage detection module was successfully deployed in Jetson TX2 that runs at a frame rate of 12 frames per second (FPS). On the other hand, the IED detector can identify IED components such as the battery, mortar, and wire. After evaluation, the IED detector achieved 77.59% mAP using Faster Region-based Convolutional Neural Network (R- CNN) in the X-ray dataset. The threat objects from the X-ray image were also analyzed using image processing techniques to get the dimension of the object and the distance from a reference object.