Research On Ground Target Recognition Technology Based On Multi-sensor

The 21 stcentury,with the rapid development of artificial intelligence,social demand for the auxiliary driving and intelligent traffic monitoring is increasing.And the ground target recognition is the key of auxiliary driving,traffic intelligent monitoring and other fields.Traditional ground target recognition system is based on the visible light sensor devices.Under the condition of light,good weather,the equipment can collect clear images.But at night,the rain,snow weather,and smoke fog cover,they are difficult to obtain the effective target information.But often the application of these extreme conditions is the need to focus on.Therefore,only using a single visible light sensors can not meet the needs.In order to solve this problem,in this paper,the integrated use of visible light and infrared sensor for identification,making the vehicle as the research object.A ground target recognition system based on multi sensor information fusion is developed,which can break the limitation of the night,smoke and other conditions,and realize the whole day work.First,in order to be able to accurately extract the target characteristics of the vehicle,this paper studies the smoothing filtering algorithm for visible light image and infrared image.To the visible light image,Gaussian template mean filter is adopted;To the infrared image,median filter is adopted.Second,In order to improve the efficiency of recognition of the moving target,this paper studies the moving target detection algorithm,makes improvements for traditional moving target detection algorithm,finally combining the inter frame difference method and background difference method,realizes the detection of moving vehicles.After detecting the vehicle target,we need to fix the position of the target area in the image.In this paper,an algorithm based on the coordinates of the center of gravity connected domain mark,and successfully locate the target area.Then the Haar-like rectangular features is used to represent the vehicle,the characteristics of the original library is expanded,adding rotation single rectangle features to describe the vehicle shadow region.Final,improve the arithmetic to solvethe problem that the training occurs to abort in the process of using traditional AdaBoost algorithm training strong classifier because of weight distortions.Increase the filtering mechanism in each round of iteration training samples,improve the performance of the algorithm,and successfully trained visible light and infrared vehicle classifier,vehicles of around the clock target recognition is realized.In this paper,recognition of visible light and infrared image target is mainly studied,and ground target recognition system is established.The experimental results show that the system has good real-time performance and robustness.Research results for machine learning techniques in areas such as auxiliary driving and intelligent traffic monitoring application has a certain reference value.