Research On The Moving Object Detection Algorithm In The Elderly Care System

The intelligent elderly care surveillance system is a new type application of computer vision technology. One of the most important technologies in the system is the identification and segmentation of the moving objects. Because the application scene of the intelligent elderly surveillance system is complex and changeable, it is difficult to extract the complete moving target. In order to detect a more accurate and complete moving object in the complex surveillance scene, this dissertation will study the image processing methods and the moving object detection algorithms.In this dissertation, we use the median filtering method, the morphological processing and the connected processing methods to achieve the prophase processing of the target detection. In order to detect and segment the moving targets, we use the Inter-Frame Difference algorithm, the Gaussian Mixture Model algorithm and the Code Book model algorithm. The Inter-Frame Difference algorithm used in the elderly monitoring system is suitable for the scene where there is obvious difference between the foreground pixels and background pixels. Because the moving object extracted by this method is not complete, it is usually used as a supplemental method for other target detection algorithms. The Gaussian Mixture Model describes pixel changes with more than one Gaussian distribution. In this dissertation, we implement the object detection algorithm with the Gaussian Mixture Model, and quantitatively analyzed the impact to detection results causing by learning rate and Gaussian distribution number. At last, we give a set of generally applicable parameter values for the Gaussian model.The basic idea of the Codebook model algorithm is to encode the pixel sampled into a Code Word within a period. This method has three problems: first, the background model cannot adapt to the slow light changes in the scene; second, after the establishment of the background model, any new background objects is classified as foreground objects; in addition, when major changes happened in the scene, the background model cannot detect moving targets. In order to solve these problems, we modified the traditional Code Book model. Using the square block of gray color space model to replace the cylindrical RGB color space used in the traditional codebook, we can solve the problem of slow light changes. We put forward a double Code Book model to solve the update problem that caused by the new background object in the scene with a main Code Book and a cache Code Book. The role of the cache code is to quantify the new objects that appear after building the main codebook. That is, to move the Code Word in the cache Code Book to the main Code Book according to the set time threshold. Also through establishing the cache Code Book when there is sudden change, we can achieve the background correction by replacing the main Codebook with the cache Code Book that satisfies a certain threshold time. The Code Word of the modified double layer Code Book model has small space occupied and the Codeword matching and updating process does not involve any complex floating point operations. Therefore, it has more efficiency. In this dissertation, we compare the modified double Code Book model and the traditional Code Book model in two aspects: the algorithm's effectiveness and real-time ability. The result has proved that the modified double Code Book model is more robust and has higher detection efficiency.