Study On Recursive Image Processing Algorithm With Variable Parameter

Small moving target. detection and classification in image sequences are important problem in computer vision and image processing. Especially, real-time detection is a key task for automatic targets reorganization system. In this paper, researches based on chrematistic of wide field watch automatic targets reorganization system are developed.In theory, a temporal recursive low pass filter model is developed and used to update the background dynamically based on multi-frames moving weighted averaging and motion detecting. Actually, several proposals lie in the derivation of the parameter k in the model. Comparison and analysis among these proposals are discussed. With the condition of constant false probability, a real-time implementation algorithm is presented based on temporal recursive low pass filtering and frame correlation detecting.In experiments, simulations of temporal-spatial filtering and spatial filtering are performed due to research noise reduction and background restrain. Simulations of small moving targets detecting algorithm based on temporal recursive low-pass filtering are also run. Several kinds of parameter k in the algorithm are analyzed. Temporal recursive low-pass filtering is an effective method in real-time detecting small target with condition of complex backgrounds. It can implement the background updating and improve signal to noise ratio. A real-time filtering system based on this kind of temporal recursive low-pass filtering algorithm is presented, which is made with video mixer chip HSP48212, static memories, in-system programming devices and embedded microcomputer. The system principle, structure and control relations are analyzed. The algorithm can enhance targets and suppress background is shown in experiment results. Adaptive threshold algorithm is applied in order to get the binary mask of the moving targets. Parameters in adaptive threshold algorithm are calculated in detail.Conclusion is proved with researches on theory and experiments: If the camera is stationary with respect to the environment, a large part of the image corresponds to static background, subject only to illumination change. The variations of the background are normally slow compared to the intensity variations caused by moving targets. It means that the temporal correlation of background is stronger than small moving targets. A temporal recursive low pass filter can be used to update the background dynamically and separate the slow process in the image sequences from fast process. Comparing a current frame with updated background (reference frame) can detect changes due to targets motion. With the condition of constant false alarm probability, applying adaptive threshold to the difference image, the binary mask containing the general shapes of the moving targets is obtained.