Personendetektion aus Tiefendaten

Research project

2015


Technologies

  • C++
  • ROS
  • Qt
T he objective of my research project in 2015 for my masters degree in Computational Visualistics at the University of Koblenz was to evaluate a method for person detection from RGB-Depth based sensors, such as the Microsoft Kinect. The research project was supervised by the Active Vision Group of the University.

Extraction a set of features from the depth point cloud that could be used for detection of persons and their hands. The features were transformed into the Fourier frequency space for preprocessing. An extracted feature vector was then used for reconstruction using an inverse Fourier transform. Together with color-spaced-based features for skin tone detection, these shape features were extracted from the Cornell Activity Dataset [CAD_60 and CAD_120] ‒ annotations of the data was performed beforehand. Using a Suppport Vector Machine, algorithms were trained on a subset of the data ‒ the rest of the annotated dataset was then used for evaluation.

Final presentation

‒ Publication
Fourier Features For Person Detection in Depth Data
Seib V., Schmidt G., Kusenbach M., Paulus D. (2015)
In: Azzopardi G., Petkov N. (eds) Computer Analysis of Images and Patterns. CAIP 2015
Lecture Notes in Computer Science, vol 9256
doi: 10.1007/978-3-319-23192-1_69