2D trash tracking for analysing robot picking (2014)

In this small work, a simple Lukas-Kanade tracking is used to track feature points in time in the context of robotic recycling system. The main idea is to answer a question: what was picked? When consistent vertical motion is detected among trash, everything moving is tracked backwards in time to register the picked element in top-view spectral camera view. This allows analysing whether the intended object was actually picked after a decision was made. Top-view spectral view was not available for this work, but picked points are tracked backwards in time.

Case1: When tracked points among trash move in vertical detection, a pick is detected and the points are tracked backwards in time.

Case2: When tracked points among trash move in vertical detection, a pick is detected and the points are tracked backwards in time.

Case3: Everything is tracked until a pick is detected.

Case4: Everything is tracked until a pick is detected.