The Study And Implementation Of Real-Time Infrared Image Matching Algorithm

Infrared image technology is one of the hotspots research in today’s society, and the infrared image matching is one of the key researches of the infrared image technology. In this context, the paper studied the infrared image matching algorithm, and implemented it in hardware. The main contents of the paper include:1. Studied the general process of the infrared image matching, and realized the image cropping, filtering, edge extraction and image binarization preprocessing.2. Studied the image matching algorithm based on grayscale: normalized correlation algorithm and grayscale coding algorithm, and simulated these two algorithms. The results show that under the premise of the grayscale linear change, these two algorithm could achieve accurate matching result, but the latter is better than the former in the matching speed. Against the large computation shortcoming of the normalized correlation algorithm, the paper proposed an improved method by using pyramid algorithm, experiments show that the improved method effectively enhanced the speed of the algorithm.3. Studied the image matching algorithm based on the frequency domain:the phase correlation algorithm and the changing algorithm based on the log-polar coordinate mapping, and simulated these two algorithms. The results show that the phase correlation algorithm could accurately calculate the relative displacement between the images, but for the images with rotation or scale change, this algorithm could not match accurately, the changing algorithm based on the log-polar coordinate mapping which turn the rotation and scale changes in the Cartesian coordinate into displacement in the log-polar coordinate is the improvement of the phase correlation algorithm. Against the large memory consumption and the big matching error while match the images with grayscale distortion of the changing algorithm based on the log-polar coordinate mapping, the paper proposed an improved method by adding edge extraction before the image do log-polar mapping, experiments show that the improved algorithm not only enhances on the computing speed and the processing effects, but also greatly reduces memory consumption.4. Built the hardware platform based on FPGA+DSP, taking the hardware platform memory capacity, the cache memory of algorithm and the real-time factors into consideration. Designed a scheme that combined the improved normalized correlation algorithm and the improved changing algorithm based on the log-polar coordinate mapping, and implemented it on the hardware. Realized the speed of33frames per second and a pixel positioning accuracy, and met the demand of practical application.