Object Tracking in Robotic Micromanipulation by Supervised Ensemble Learning Classifier

Autonomous micromanipulation has a great potential to impact every research field concerning objects at a small scale. In this paper, we report our work on detection and tracking of a transparent SU-8 microchip in 3D Cartesian space during micromanipulation. Conditions such as occlusion, variant object orientation and poor edge prominence hinder the implementation of conventional vision algorithms. To enable tracking in such difficult conditions, an object detection classifier that utilizes an ensemble machine learning algorithm has been implemented. The classifier has been trained with 165 and 85 unique positive and negative samples, respectively. Object detection was achieved at a distance of three times the nominal depth of field with maximum tracking error of only 12 % of the object size.