No abstract

The purpose of our demo is to show the application and performance of some low-complexity image descriptors in object recognition under realistic circumstances. We built a client-server system where several image retrieval methods and image segmentation approaches can be tested with the help of a network connected Android device (mobile phone, table or head mounted computer). A modified version of the CEDD (Color and Edge Directivity Descriptor) is proposed, as the most robust lightweight descriptor found in our tests, and manual or saliency based object selection are also included. The main purpose of the demo is to show the possibilities of lightweight object recognition with the modified descriptor and different object segmentation.

No abstract

No abstract

-In this paper we present a new motion recognition algorithm based on skeleton and depth map data extraction from two generations of the Kinect sensors as a first step to support manufacturing process optimization. A real production line activity was simulated in a laboratory environment and a specific motion -the barcode scanning -was recognized and validated by semi-automated log file processing. We show that the proposed methods give appropriate accuracy by using whichever Kinect sensor.