Ground Maneuvering Target Recognition Based On Far-infrared Thermograph

With the increasing social demand for the traffic monitoring and automotive driver assistance, there is a growing demand for intelligent target recognition system. However, at the night, fog,smoke and other enveloped, the common visible light imaging device cannot get effective information from the set of scenarios. That seriously hampered the development of these intelligent. But the development of infrared technology is gradually overcome this bottleneck, such as far infrared thermal imager. It is use the infrared detector to receive the thermal radiation of the measured target to imaging. It can convert the infrared radiation spectrum which the human’s eye cannot perceive into the visible spectrum. It broken the limitation of the dark night, and it has a certain amount of smoke penetration ability. It is not sensitive to illumination changes, which can realize remote all-weather observation. It created a possible for which use image processing technology to finish the detection and recognition of the ground maneuvering targets.Based on the infrared thermal imager, this paper made a detailed analysis which includes the infrared image processing, the maneuvering target real time detection, the target feature extraction and selection, the target recognition. In the infrared image preprocessing stage. Firstly analyzes the characteristics of infrared image, according to its characteristics to choose the appropriate image enhancement algorithm to enhance the infrared image. The median filtering is a non parameter estimation algorithm, alternative process pixels exist blindness and randomness. It will lead the image edge and details have some extent broken. This paper made an improved median filtering to save the edge details. In the maneuvering target detection stage, because of the traditional maneuvering target detection algorithm cannot quickly complete the detection of target. Based on the original algorithm, this paper uses the description of adaptive change background pixel number of Gauss distribution to reduce background modeling time. It also can improved the running speed and put forward an method which combined the background difference method with the frame difference method to complete extraction the maneuvering target. In the target feature extraction and selection stage, in order to improve the execution efficiency of feature extraction, this paper puts forward a multiple target localization algorithm which based on the grey value projection. It completed the multiple targets location which appeared in the image. Finally finished the detection and recognition of the moving target by through the maneuvering target recognition system based on SVM, and did a analyzed experimentally to the recognition correct rate. The results show that the system has certain practical value and research significance.