Comparing Investigation of Images and 3D Mesh Model For Categorizing Electric Appliance Components

Knowing about the type of objects that exist in real world is important thing for automation of conveyance or classification task. This paper reports our challenges the purpose of which is to categorize objects that are incorporated into home electric appliances. Our target includes small sized objects, flexible objects, and objects that have various appearances by the difference in view-points. Sensing methods we used are two types of measurement data; (1) Mesh model measured by a 3D digitizer, and (2) multi-viewpoint images captured by a high-resolution camera. Using these sensor data, we studied about feature descriptions. These descriptions are designed by regarding the characteristics of target objects and the property of each measurement data. Figure 1 shows our target objects that are composed of 31 series of electric appliance components. They include small sized objects (e.g. screw and clasp), relatively large objects (e.g. electrical circuit sized hundreds millimeters), shiny and transparent objects (e.g. glass), and flexible objects (e.g. cable and wire). Meanwhile, some of components have a big difference on their appearances or shapes even though they are grouped as a same category. The goal of this research is to design high-precision classifier for these 31 categorized objects.