Edinburgh Napier University

  Infrared pedestrian detection systems struggle with real-time processing performance. Known solutions are limited to either low resolution systems with basic functionality running at full frame rate, or software based techniques featuring higher detection rates with full set of features, however running only in off-line mode for statistical analysis. Here, a comprehensive solution for real-time pedestrian detection is described. This research project includes investigation of possible solutions, design, development and implementation of a pedestrian detection system, processing data from infrared video source in real-time. Design requirements include processing at full frame rate as well as low memory and system resource consumption. The memory utilization is one of the major concerns since high demand for memory resources is a critical aspect in most image processing applications. For the purpose of this task, a number of general purpose image processing techniques were revised, taking into consideration the suitability for infrared pedestrian detection. These tasks include background separation, acquisition noise removal and object detection through connected component labelling. They are discussed and addressed in individual chapters. Various techniques for background segmentation are discussed. A chronological review of popular techniques is provided. The proposed architecture for background subtraction is based on selective running average for adaptive background model, supported by adaptive thresholding based on histogram calculation. In order to remove acquisition noise, a dual decomposed architecture was introduced, based on mathematical morphology and basic set theory definitions. It includes both erosion and dilation performed in a pipeline. For the purpose of object detection and feature extraction, a connected component labelling technique was employed, based on a single pass approach to fulfil real-time processing requirement. The system was implemented, verified and tested on XUP FPGA Development Board with Virtex-II Pro XC2VP30 chip from Xilinx. Details and limitation of the specific implementation are discussed. An overview of experimental pedestrian detection results is provided. The thesis concludes with system analysis and suggestions for future work.