Multi-motion Object Detection And Recognition In Video Sequences

Multiple object detection and recognition is a hot spot in computer vision and digital image processing research, widely used in security monitoring, information management, intelligent operation etc.. In this paper, the study of the impact of complex illumination change on multiple moving objects detection, use the ratio of gray information to avoid the illumination change of background models under multi-light scene; Combining texture information with local information reduce the illumination changes and noise effect on the performance of object recognition; Fusioning color attribute and variable characteristics of the classifier, significantly improve the object recognition accuracy under the interference, such as partial occlusions, distortion, smallscale objects recognition. The main work of this paper include:First, using the gray level ratio information cascade Gaussian mixture model, improve the reality of moving object detection accuracy in multi-light scene, avoiding the influence of light on the image foreground and background pixels.This paper compares the existing moving object detection method, builds the mutli- illumination model in realistic scene, Uses this model represents the impact of changes in illumination of pixels on the image, using the grayscale ratio information correct the light change. Finally using the Gaussian mixture model cascading the gray scale ratio information realize moving objects detection, response to the illumination changes.Second, combining the texture information with the characteristics of the local structure to reduce the effects of illumination changes and noise, improve performance of object recognition. This paper analyzes the existing local texture feature and extraction method, using the Lab color space to avoid the gray space information loss, combining the Lab color space with the local average histogram model describe the color and brightness of image information, and then using the characteristics of the fusion of color and the local structure information to realize reliable object recognition.Third, using the variable feature classifier constituted with color attribute, tonal and local location information, weak the interferences of the local shelter, deformation of object, small-scale objects, improve the accuracy of object recognition. This paper uses the way of weighted fusion tone color attribute information, to build image color features and describe the object state information. Using the way of object split added the component’s location information in the local characteristics to build the variable model avoiding the local structure feature detection error. Using the mixed feature of color features and variable model achieve the high robustness object recognition.Finally, using the grayscale ratio information detect the moving objects, and using the local structure information, color attributes, texture and the variable model to distinguish the objects, the experiments testify that combining color attributes with the variable model has more reliable performance.