Real time target tracking algorithm

Polytech’Clermont-Ferrand, a French Engineering school, ran a university project in partnership with Sundance. The main work was focused on building a standalone embedded system based on the Sundance advanced video system: SMT8039.

This system is able to detect a person or an object in a public area from a video camera input. The application executes a tracking algorithm in real time. The targets are identified by a cross and a rectangular frame. The processing is implemented on the Texas Instruments DM642 DSP and the Xilinx Virtex-4 FX60 FPGA.

The embedded application was designed with the 3L Diamond software to give best performance. The algorithm is split into four DSP software tasks and the frame-grabbing is done in the FPGA firmware:

  • Acquire the current image and store it into a buffer,
  • Subtract every pixel of the current image from every pixel of the first image (stored at the beginning of the process),
  • Compare each result with a threshold to get a black and white image. So the moving objects will be illustrated in black.
  • Calculate the barycentre of all the black points, and display the result on a console screen with a cross and a rectangle around the person.

Typical video applications illustrated by this concept are:

  • Human and object detection, recognition and identification,
  • Person or object counting,
  • Traffic monitoring,
  • Security and Surveillance IP Cameras…

Polytech’Clermont – CUST, Blaise Pascal University, Clermont-Ferrand, France