Next Best View Estimation With Eye In Hand Camera

摘要: For many robotic systems, automated data acquisition is an important part of detecting and recognizing objects. This true especially for robots like the PR2 which are supposed to operate in human environments our homes. Human living usually highly cluttered. Thus, oftentimes, a scene can not be fully observed from one position only. scenes this, treated as sensor placement problem, also known Next-Best-View (NBV) problem. The NBV approach seeks single additional order improve knowledge environment. However, set possible positions constrained by abilities. instance, if mounted on head robot, only changed driving robot around. In kitchens this make information gathering difficult, countertops accessible side, making it difficult observe objects countertop different angle. We address automatic scenarios work. To enable handle positioning settings we equip with camera system high dimensional manipulator. using Microsoft Kinect, provides us point cloud representation scene. increase degrees freedom allows choose bigger viewing allowing area would otherwise inaccessible. Additionally, consider possibility planning trajectory such that movement used collect even more data. following section introduce main ideas approach. Then illustrate discuss first results obtained through