Research On Algorithm Of Moving Object Detection And Gait Recognition In Video Sequences

Due to the complexity and unpredictability of the monitoring environment, there still is false detection existing in moving target detection algorithm and it will have effect on target tracking and identification; at same time, compared with the identification algorithm with gait characteristics, fingerprint, iris, etc have long-range and non-contact features; but currently there is not high recognition rate in the gait characteristics identification algorithm. In order to meet a higher requirements of monitoring system, this paper will conduct a in-depth research in the following three aspects:Firstly, for the deviation of the mean Gaussian mixture model only for a single pixel independent processing and deviation into a small and easy to cause erroneous detection by using the pixel correlation between the characteristics, the use of the neighborhood does not match the pixel point further matching Classifying; threshold and automatically adjust the range of the image pixel value according to different times. Experimental results show that the improved algorithm can effectively resist the noise to reduce the misjudgment of the background pixels and improve the prospects for target detection effect.Researching on the nature of the geometric moments, complex moments, and orthogonal moments, it gives a different moment invariant feature function. Using the function to extract the original image and the image geometric transformation invariant features, then analysis different characteristic function for image description capability, noise immunity and redundancy.In order to effectively improve the recognition rate of gait recognition algorithm, it is proposed a transformational gait recognition algorithm based on Radon transform and the analytic Fourier-Mellin. The algorithm change separately from the original image rotation and scale changes to the phase change and amplitude change, after the same function to extract the target image by defining rotation and scale invariant features for classification and identification. As eliminating the need for a target image binarization and normalization, it can keep the image for more details and avoids the error of resampling and weight. Experimental results show that the gait recognition algorithm is applied to a higher recognition rate and can achieve better recognition results.The content of this study can solve the moving target detection errors, then effectively improve the gait recognition rate. It has certain application value and significance for target detection system of video surveillance and identity recognition.