Application Research On The Technology Of Bird Identification And Tracking Under Driving-bird Monitoring System

In recent years, there are more and more reports about bird harm in commercial crop production in China, not only the outdoor cultured commercial crops but also crops inside the greenhouse often suffer from the bird harm and attack. At present,after the analysis of various bird repellent methods, it is found that the ultrasonic has good effect in repelling birds, but which requires voice for a long time and also requires a lot of power, thus resulting in the waste of resources.Therefore, it is necessary for us to add bird detection and tracking methods as the ultrasonic bird repellent switch. Presently, bird detection methods at home and abroad mainly include the radar method, thermal imaging method and the recognition method based on image processing technology. The first two kinds of detection methods are low in the cost performance, which often require expensive equipment, and the farmers are hard to bear such a high expense.The detection method based on image processing usually only needs a camera and a small processor, which is high in the cost performance for farmers who are nearly unbearable to birds harm and attack forsuccessive years, thus it is easy to be popularized.With the screening and data analysis towards 33,623 samples on ImageNET, in this paper, 31,495 positive bird samples and 1,064 negative bird samples are obtained.Training process of the bird model.Various common bird image data is obtained by ImageNet. Process the picture cutting on the original image, manually classify those pictures with bird information, conduct training towards initial data samples with the machine learning method, extract feature vectors of these data by HOG feature and also of, and carries on thewindow width weighting and optimized the model according to results of the training.Recognition process.Preprocess the input image by using the Laplacian Gaussian function(Gaussian function can eliminate image noise and smooth images; Laplace can properly sharpen images on the basis of smoothness and enhance the original details).Then operate the picturepartition towards the input image(20 x 20), and also carry on HOG feature processing and window width weighting on the pictures, recognize by using the feature model trained in the process of bird model training, and finally track the bird videos by MeanShift algorithm.Then, in 4 normal MP4 format video files as experimental object, under the environment of VS2010 tracking target test. For the experiment of each video can track the target, prove the feasibility of the algorithm. Finally, according to previous research, a series of system analysis, system and background management system to design the front-end, eventually get Intelligent Bird Repellent system design scheme of the system.