Research On Detection And Recognition Of Moving Objects In Intelligent Video Surveillance

With the continuous advancement of science and technology,video surveillance has been widely used in banking,power,transportation,safety supervision and other fields,and the demand for video surveillance has also increased dramatically worldwide.However,current surveillance video is mainly used for artificial real-time monitoring and post-search.At the same time,many valuable information in video surveillance are also covered over time.These situations have caused huge waste of human resources and data resources.Therefore,it is very necessary to study computer vision.Detection and recognition of moving targets is an important branch in computer vision.Vibe algorithm is a typical moving target detection algorithm.However,this method has the disadvantages of slow elimination of "ghosts" and poor immunity to global light changes.In image classification,VGG16 is a classic image classification network based on convolutional neural network,and it has excellent performance in various image classifications.However,VGG16 also has disadvantages such as large amount of parameters,slow training speed,and easy overfitting.In order to solve the above problems,this paper first improves the rumbling algorithm,effectively improves the rumbling algorithm,effectively improves the "ghosting",poor adaptability of light mutations and other issues.Secondly,this paper adjusts the VGG16 network,which effectively improves the network model’s large number of parameters and slow classification speeds.Finally,this paper combines two methods to realize the detection and recognition of moving targets.The main work of this paper is as follows:(1)The combined Vibe algorithm and three-frame difference method proposed an improved moving target detection algorithm.Simultaneously,several data sets such as Pets2006 and wallflower were used to test and verify the proposed algorithm.(2)On the basis of VGG16,four image classification models based on convolutional neural network are proposed by modifying the size of the input image,adjusting the number of neurons in the full-connection layer,and introducing a global average pooling layer.The CIFAR-10 data set is used.Several models were evaluated for accuracy and classification speed.At the same time,we also use the self-built data set to train the model and provide a classification model for the detection and recognition of subsequent moving targets.(3)Combine the proposed moving target detection algorithm with the image classification model based on convolutional neural network to implement a complete set of moving target detection and recognition procedures,and use different size picture sequences to test the classification performance and effectiveness of the program.This article aims at the moving target detection and the image classification technology,this paper proposes a method combining the improved Vibe algorithm and the VGG16 moving target detection and recognition method,and carries on the programming realization.The program can detect and identify moving objects in the video,effectively reducing the labor costs in video surveillance.