Detection And Extraction Of Complex Background Moving Objects Based On Embedded Platform

A key component of the intelligent video analysis system is moving target detection,which has been gradually promoted and applied in many fields such as security and transportation.However,the traditional PC-based moving target detection system has limitations,such as high hardware requirements,large volume power consumption and poor real-time performance,which cannot be applied on a large scale.With the development of video processing and embedded related technologies,solutions have been found for the above problems and difficulties.The moving target detection system based on embedded DSP utilizes the features of DSP chip,such as powerful digital signal processing performance,low price and easy to integrate application.It is a subject with great research value and significance.This paper studies the fractal theory and moving target detection algorithm.The main purpose is to improve the accuracy and real-time performance of moving target extraction in complex backgrounds.A target detection algorithm combining color and fractal features and an improved Vi Be algorithm are proposed to achieve movement in complex backgrounds.The above algorithm is verified by experimental video testing and quantitative data analysis,and complete the implementation of the algorithm on the embedded platform.The main work of this article is as follows:Firstly,it deeply researched the fractal theory,and made use of the difference between the natural scene and the artificial target in the fractal feature,and proposed the target detection combining color and fractal feature.First convert the RGB color space to Lab color space,then use improved K-means clustering to remove large background areas,and finally calculate the regional fractal dimension and fractal fitting error.The combination of the two fractal features can accurately exclude small areas of background singular areas interference,detecting multiple targets in the image to be tested.The method of complementary information effectively combines the dual advantages of image color space and gray space,and improves the detection speed and accuracy of the algorithm.Secondly,this paper studies and analyzes the principle,advantages and disadvantages of vibe algorithm,including background difference,frame difference and vibe algorithm,and improves the vibe algorithm.First,when building the background model,extract 8 neighboring pixels corresponding to the pixels of the first three frames of the detection video to establish a background sample set,which expands the selection range of sample points,which can effectively reduce the repeated selection of sample points to improve the classification accuracy.In order to adapt to the background disturbance,the adaptive update threshold is adopted,and the background and foreground pixels are more accurately divided after the adaptive matching radius.Finally,the multiple features of the image are integrated to further classify the extracted motion area and remove the interference of the background area,improve the accuracy of moving target detection,and simulation verification algorithm and comparative analysis.Finally,the system selects the ADSP-BF609 processor to build a moving target detection system.According to the system process and the required functions,it is refined to each module to achieve it.The functions module design plan of video acquisition,preprocessing,external storage and video output are given.The control register of the OV9650 camera is initialized through the TWI interface on the ADSP-BF609 board to achieve image quality control.The image acquisition and storage are completed through the parameter configuration of the parallel expansion interface EPPI1 and its corresponding DMA controller.Through the PIXC the parameter configuration of the processing unit realizes the conversion of the image format from YUV to RGB.The video encoding chip(CH7024)completes the video encoding output,and outputs the detection result of the moving target through the display simulated by the device.Completed a moving target detection system that integrates video acquisition,algorithm processing,and result output.It takes advantage of the embedded platform and provides a feasible solution for video surveillance in unattended locations.