Research On Human Behavior Recognition Algorithm Based On Skeletal Joints

Image segmentation aims to classifying pixels with the same or similar properties in the image into non-interference sub-regions(pixel sets)with unique properties,so that features in the same sub-region have certain similarities,and different regions have certain differences.Image segmentation is an important branch of contour detection whose results can be characterized as a closed contour curve.Contour detection algorithm can also transform the contour result into a closed segmentation image through some special techniques such as watershed algorithm.As an important image processing method,image segmentation technology cannot only divide the image into different objects,helping the computer to understand the image content through object locating,but also make the processing object independent of the image background,ensuring that it is not affected by background factors in subsequent processing.In practical wide-ranging applications,image segmentation techniques are often combined with domain-specific expertise to handle different types of problems,such as,the detection of pedestrians,faces and even tumors in medical images,the problem of image understanding in machine vision,object segmentation and motion localization in traffic surveillance video.Among these problems,the image segmentation technique is more often used as a preprocessing means to obtain a region or object of interest to achieve the purpose of reducing the computational speed.As an important image processing method,image segmentation algorithm is an important topic in the field of computer vision.It has been widely concerned and has published many research results in the top conferences and journals of computer vision.From the time when computer images appeared,image segmentation technology began to develop along with image applications.From the most basic threshold segmentation method in 1975,to traditional image segmentation algorithms such as clustering and graph cuts,to the most popular neural network-based semantic segmentation,after decades of development of image segmentation a large number of algorithms have been generated.These algorithms can be roughly classified into texture information-based segmentation algorithms and semantic information-based segmentation algorithms.The former are traditional image segmentation techniques that use image texture features to classify pixels,and the latter are to extract high-level semantic feature through neural networks.The pixels are then classified into predefined labels.In the second chapter,we will also analyze the algorithms in these two directions.After analyzing the current algorithms in the field of image segmentation,this paper proposes a novel contour detection and image segmentation algorithm based on spatial color density distribution.Our algorithm firstly evaluates the kernel function by estimating the probability density function of the 5-dimensional color space of the image to obtain the distribution information of the image pixels in the 5-dimensional space,and then uses the maximum-value mapping to map the 5-dimensional spatial function values and the corresponding color values onto 2-dimensional density images and color images.The density image is a good texture feature,which contains the distribution information of the objects in the image and can represent the pixel texture information.Therefore,after the density image is calculated,the contour information in the density image is extracted by the non-maximum suppression method.In addition,the combination of the density image and the color image can well represent the texture information of the pixel.After defining the distance of the pixel,the algorithm clusters the pixels by DPC algorithm,and the pixels are divided to achieve image segmentation.Since the calculation of the color space density function estimation is too large,in our experiments,algorithm on GPU needs to be realized by CUDA programming.The main works of this paper are as follows:1.The existing image segmentation algorithms based on texture information and semantic information are analyzed and studied.The typical algorithms are described.Through the analysis and research of existing algorithms,the understanding of image segmentation problems is deepened.2.After analyzing and researching existing image segmentation techniques,a contour detection and image segmentation algorithm based on spatial color density distribution is proposed.When the algorithm is implemented,GPU acceleration through CUDA programming overcomes the problem of large computations when estimating space probability density function.The algorithm consists of two partial algorithms,which are the contour detection algorithm based on non-maximum suppression and the image segmentation algorithm based on DPC clustering.The former outputs the contour intensity detected in the image,and the latter outputs the segmentation results at different levels.3.The algorithm proposed in this paper performs contour detection and image segmentation experiments on the BSDS500 database.Experiments verify the effectiveness of the contour detection and image segmentation algorithms proposed in this paper.