Research On Target Detection Algorithm Based On SVM And RPCA

Target detection is an important research direction in the field of computer vision research.It is a prerequisite for other advanced visual tasks.The target detection algorithm based on Support Vector Machine(SVM)is the mainstream algorithm of machine learning.Robust Principal Component Analysis(RPCA)can remove the noise pollution in the image to be detected to some extent,which is very helpful to improve the detection rate of target detection.In this paper,the support vector machine is added to the saliency and convolution respectively.The neural network performs target detection,and combines robust principal component analysis and mixed Gaussian model to obtain a new algorithm for moving target detection.To some extent,it solves the problem of low robustness and large computational complexity of target detection algorithm research..The main research work is as follows:(1)A deformable component SVM target detection Deformable Part Model(DPM)algorithm based on saliency analysis is proposed.On the basis of the SVM target detection model of the deformable component,a significant detection model is added,a significant feature map is generated by using the global contrast-based saliency analysis method,and the salient feature map is detected by the deformable component SVM model.Experiments show that the algorithm reduces the amount of calculation,the detection time decreases,and the detection rate increases.(2)A pedestrian detection algorithm based on support vector machine for convolutional neural networks is proposed.On the basis of establishing multi-scale image sub-blocks on the target image,the feature extraction and classification of the image is performed by the classical network LeNet-5 of convolutional neural network,and then the image sub-blocks identified as pedestrians by the convolutional neural network are processed by SVM.Sub-category.Experiments were carried out on the INRIA pedestrian dataset.The experimental results showed that the pedestrian detection accuracy and recall rate were significantly improved.(3)A moving target detection algorithm based on RPCA and hybrid Gaussian model is proposed.In the actual scene,video frames often have noise pollution such as illumination and occlusion.In order to reduce noise pollution,it is proposed to perform robust principal component analysis on video targets based on the research of moving target detection,which solves lighting and so on to some extent.The effect of noise on the detection effect effectively removes the effects of noise.Experiments show that the proposed method shows good robustness to moving target detection.