• Thomas Ægidiussen Jensen
  • Henrik Anker Rasmussen
This project deals with the problem of automatically generating statistics on the movement of people in complex indoor scenes. These statistics are to be used for commercial purposes, e.g. to determine rental prices of shops, placement of ads, etc. Commercial solutions for this exist today, ranging from simple solutions like infrared beams to advanced and expensive solutions like thermal sensors. However, common for these are that they require extra hardware to be installed. Furthermore, existing solutions that are able to generate advanced statistics on the movement of people require expensive hardware.
In this project, a system is developed that automatically generates statistics on the movement of people using footage from surveillance cameras in the Friis shopping mall. Thus, the system developed in this project requires no installation of extra sensor hardware. Furthermore, it requires no prior knowledge about the scene.
The system consists mainly of a multi-human tracker and a visualizer. Prior to generating the actual statistics, the system estimates information about the scene on a long video sequence to improve the performance of the multi-human tracker.
The system is evaluated quantitavely and qualitatively on video from the Friis shopping mall and on video recorded for this work at Aalborg University. The results show that while a higher error is measured in the statistics generated on video from the Friis shopping mall, the generated statistics generally resemble the actual statistics. Thus, it is seen that the system is able to capture the trends of the movement of people and make a visual representation of these.
This project makes the following contribution to the problem of tracking multiple persons in complex scenes: Enhancing the output from human detectors by exploiting scene information automatically generated from reliable human detections. Also, estimated trajectories are used for generating statistics on the movement of people. Lastly, suggestions on possible improvements and extensions of the system are given.
Publication date31 May 2011
ID: 52685250