Study On Human Recognition Technologies Of Infrared Images Based On The Micro UAV

Recent years, with the high development of infrared technology, the human recognition technology in infrared images has been used widely in traffic management, monitor, driver assistance system, and so on. It has a wide range of applications in civil field, and also draws more and more attention in military field. With the rapid development and the application of Unmanned Aerial Vehicles, the human recognition technology in infrared images that applied in high altitude remote sensing platform has gradually become a research hotspot.Since the micro UAV is a platform with high speed movement and high off the ground, the human in the infrared images are smaller, the background is complicated and the noise is larger. According to the characteristics of the infrared images taken from high altitude, a infrared human recognition system is designed. The key steps of the system include: image segmentation, candidate determination and feature extraction. The main work is as follows:1. In order to improve the accuracy of image segmentation and reduce the calculation, a double threshold image segmentation method is used in this paper, using global threshold and local threshold to achieve the segmentation of the target areas. In the algorithm, the statistical variance method is employed to get the global threshold, which is used for initial segmentation. Then a cross sliding window is used to scan the processed image and compute the local threshold, which can be used to segment the infrared image accurately.2. In order to reduce the calculation amount of the feature extraction, classification and recognition, the paper determines the candidates according to the basic characteristics of human body proportion and employing the candidates areas ratio and the symmetry of vertical edge feature, which can effectively remove part of non-human target and improve the accuracy of the recognition.3. According to the basic features of the human in the infrared images, Histogram of Oriented Gradients (HOG) feature is selected to descripe the target areas. For the problem that the dimensions of the characteristic vector are too large, the detection windows are divided in proportion, which can effectively reduce the dimensions of the characteristic vector and the calculation amount.The infrared images in the standard image databases taken from high altitude are used to demonstrate the algorithms, and the result proves that the proposed algorithms can obtain satisfied effect in segmentation and recognition.