Experimental setup for object recognition and tracking

For the evaluation of novel imaging sensors, hardware architectures and image processing algorithms we are providing a real-time object recognition and tracking system. The modular design of the systems allows for an easy exchange and upgrade of the components. The system is also used for the development of new implementation concepts. One example is an multi-object tracker utilizing the dynamic, partial reconfiguration capability of several FPGAs. We provide system adaptions to use it in outdoor/maritime environments.

 

The experimental setup for object recognition and object tracking is used for the investigation of new imaging sensors, hardware architectures and image processing algorithms. The main components of the system are

 

• A pan-tilt unit
• Different cameras for visual and infrared wave bands
• FPGA boards for camera readout and preprocessing
• DSP boards for complex image processing algorithms
• A PC for user interactions, live image display and data acquisition

The modular system design allows for a fast exchange and simple update of components.

 

The pan-tilt unit is used to automatically follow tracked objects with the camera.

 

Using an FPGA for the direct readout of the camera raw data, the system supports proprietary and standard camera interfaces and the use of a DVI interface for simulated camera data. One advantage of this approach is the minimized latency for the data processing. Another advantage is the possibility to directly apply custom processing steps like non-uniformity corrections on the camera raw data. The FPGA is also used for the implementation of simple object tracking algorithms.

 

A floating-point DSP is used for more complex algorithms.

 

For the hardware and software implementation we use the standard tools.     

 

The system is available in different variants. Robust versions can be used in outdoor field trials, even in maritime environments in measurement campaigns.

 

 

 Use of the system in different environments.

 

A highlight is the use of dynamic, partial reconfiguration of FPGAs for a flexible system which is able to independently track different objects or track an object with different independent algorithms simultaneously. The tracking algorithms can be substituted during tracking at runtime without affecting the rest of the system on the FPGA. It is possible to adapt the tracking process depending on the tracking environment as fast as possible. The use of our custom reconfiguration controller and dynamic module addressing scheme makes it possible to reconfigure in a few milliseconds.

 

 

 

Schematic of a dynamically, partially reconfigurable multi-object tracker. The tracking modules can be substituded during runtime. 

 

 

References

 

Matthias Rümmele-Werner, Thomas Perschke, Lars Braun, Michael Hübner, Jürgen Becker: "A FPGA based fast runtime reconfigurable real-time Multi-Object-Tracker", ISCAS 2011 IEEE International Symposium on Circuits and Systems, Rio de Janeiro, Brazil, 15-18 May 2011.