Infrared Target Detection And Recognition In Complex Background

The target detection and recognition system based on infrared imaging has broad application prospects in the disaster relief and other fields.It has become one of the hot spots in computer vision and pattern recognition.This thesis' study focuses on the disaster relief in infrared target detection and identification which is based on infrared camera and computer.To study and solve the detection system of real-time and accuracy problems,it involves complex background noise suppression,region of interest extraction,human target detection and identification and GPU for image and video acceleration technology.The main research work is as follows:Firstly,we introduce the working principle and features of infrared imaging system and the classification of infrared camera and the feature of infrared image.Through the analysis of the infrared image characteristics and study on the noise suppression,we proposed a contrast enhancement method based on eliminating nonlinear temperature distribution,which make the contrast of target and background more distinct.And we give a detailed description of the process of the algorithm with the infrared image database of infrared target image sequence.Experiments show that this method can improve the SNR of infrared images and reduce the complexity of post-processing,and it can help to guarantee the accuracy of the follow-up detection.Then,we combined dynamic dual threshold algorithm with histogram of oriented gradients and support vector machine algorithm.It can significantly reduce the detection area of the image by dividing a candidate region.And it not only reduces the scanning range,but also avoids the according to the characteristics of the scanning window.Experiments show that the method can achieve the recognition of target images well,and compared with the original method,it has 4 times higher in real time.Again,because it has a bad real time capability of target detection and recognition in CPU platform,we transplant the histogram of oriented gradient extraction algorithm to the GPU for parallel implementation.The method can improve the speed up to 12?20 times,by comparing the CPU and GPU computing time,the analysis of the experimental results.We can conclude that it can signifthrough icantly improve the system in real time based on GPU video acceleration.Finally,we summarize the main research work of this paper,and prospect the problems encountered in this paper as well as the further research.