Papers by Keyword: Mean-Shift

Abstract: Aiming at the limitations of the traditional mean shift, such as invariable kernel bandwidth, an improved tracking algorithm with the following strategies is proposed. The target model and the candidate are described by the similarity between them is evaluated by Bhattacharyya coefficient. This algorithm firstly calculates the Bhattacharyya coefficient of the template target histogram and template background histogram and calculates the Bhattacharyya coefficient of the candidate target histogram of the current frame and template background histogram when tracking. Then judge the change tendency of the target by comparing the two coefficients and correct the tracking window width with plus or minus 10% increment for subsequent frames target tracking. The experiments prove that the present method has better stability and robustness than the traditional algorithm and the kernel bandwidth can adapt to the size change of the target.

Abstract: Mean Shift algorithm basedon floating point arithmetic calculation is very difficult to be implemented onreal-time systems. Another problem is that the tracking window of Mean Shift isnot adaptive. In order to solve the problems above, an improved Mean Shiftwhich can be implemented on FPGA is presented. Firstly, Hardware/softwarepartitioning based on the complexity of the algorithm is determined theimplementation of SOPC. Secondly, A pipeline structure is employed for MeanShift algorithm to calculate and accumulate kernel function, shift vector andtarget area in a single clock cycle, and accuracy is above 99 percent. Finally,as floating point arithmetic is more time-consuming, custom floating-pointinstruction is added into CPU to improve the algorithm operation speed.Experimental results show that the presented method can adaptively updatetarget windows and track target in 4ms stably, which improve the efficiencyof program by 5 times.

Abstract: Target tracking algorithm mean-shift and kalman filter does well in tracking target. However, mean-shift algorithm may not do well in tracking the target which the size of target is changing gradually. Although some scholars put forward by 10% of the positive and negative incremental to scale adaptive,the algorithm can not be applied to track the target which gradually becomes bigger. In this paper, we propose registration corners of the target of the two adjacent frames, then calculate the distance ratio of registration corners.Use the distance ratio to determine the target becomes larger or smaller. The experimental results demonstrate that the proposed method performs better compared with the recent algorithms.

Abstract: The Mean-Shift algorithm has very good tracking effect when the background is in a simple; but for a complex environment, tracking effect is not very ideal. Therefore, a new gray feature modeling method is proposed in this paper. Firstly, target in the tracking window is uniformly divided into even pieces. Then the pixel gray value of each block is calculated with subtraction of certain rules. Finally, the gray value of gray difference and the whole object value fusion are fused and established the object model. The object model that established not only contains the whole gray value information, but also contains the gray value differences between regions, has a more accurate description of the target, and then distinguish target from background better. The experiment results show that: the target model using the method in this paper to track based on the Mean-Shift algorithm, has good adaptability when the target is partially occluded and has better robustness for complex background.

Abstract: When the target and background in the high speed change, moving target detection. The traditional easily lost, not accurate. This paper presents a variable background frame difference method, and makes use of the MeanShift tracking algorithm simulation application. The method can detect moving objects in complex environment, and real-time tracking, can quickly and accurately detect and track when the background and target are scale, rotation, no rules of large displacement changes.

Abstract: That applying computer vision technology into the water surface floater intelligent monitoring is a creative and interesting topic. In this article, we propose the water surface floater clustering based on Mean Shift algorithm. Adopting RGB space color information, the image is divided into fixed size 3*3 small blocks, the block is mapped as vertex of graph, and each block pixel mean is mapped as pixel values of vertex; before using the Mean Shift clustering put forward carrying on one or more Mean Shift filtering smoothing process. Finally, each vertex of clustering is mapped to a 3*3 small blocks, block pixel values are vertex pixel values, and according to the proportion of all kinds of floater in the perceptual area, calculate its pollution degree, it is used to measure or evaluate the pollution index of water surface. It is showed that the method is effective and feasible by experiment.

Abstract: Image processing-based vehicle recognition is one of the important research fields in ITS. The existing methods are all based on license plate recognition and car shape recognition. Their common problem is algorithm stability. And the license plates are easy to be changed. All information about vehicles should be used to recognize them reliably. A problem to be solved is to find a method to recognize vehicles besides license plate recognition and vehicle model recognition. Vehicle license plate location and character segmentation are critical steps in the license plate recognition system, and yet there are difficult problems to be solved. Kernel density estimation and Mean Shift theory

Abstract: A novel visual tracking algorithm based on particle filter with multi-algorithm fusion is proposed. Mean shift is employed to make particles distribute more reasonably in order to maintain tracking accuracy by using fewer particles, and the genetic evolution ideas is introduced to increase the diversity of samples by applying selection, crossover and mutation operator to achieve particles resampling. The experiments show that the tracking performance of the proposed method, compared with Mean Shift Embedded Particle Filter (MSEPF), is significantly improved.

Abstract: This article describes the use of the average population density estimation methods based on video statistical analysis, and mainly discussed the research and application of the air conditioning energy-efficient system in the subway. The distributed intelligent control system in the subway station platform captured video images by more than one camera sensors, according to the computer image processing methods, for example it have the unique advantages for the fuzzy neural network to model the human nervous system in fuzzy information processing. This article used the improved Meanshift algorithm based on pixel energy to capture the moving target in the video. This method can legitimately divide the crowd by achieving the image intelligent analysis data, and whats more, it can help to get the estimation of population density.

Abstract: Object tracking has always been a hot issue in vision application, its application area include video surveillance, human-machine, virtual reality and so on. In this paper, we introduce the Mean shift tracking algorithm, which is a kind of important no parameters estimation method, then we evaluate the tracking performance of Mean shift algorithm on different video sequences.