Research Of Identification For Leaf Occlusion In Surveillance Video

This thesis was focus on the detection of leaf occlusion in surveillance video.Because of the monitoring environment in China, leaf occlusion has been the veryfrequently abnormal event in all the abnormal videos. It has a great influence on thenormal operation of the video monitoring system. This thesis proposed some methodsfor detecting leaf occlusion by analyzing the characteristics of the occlusion videos.Mainly includes the following three aspects:In the light of the characteristics of the leaf occlusion videos, combining thecommonly used target segmentation algorithms, this thesis proposed a leaf segmentationalgorithm by combining accumulated frame subtraction method and thresholdsegmentation. According to the characteristics of the moving leaf, this algorithmaccumulated the frame subtraction and segmented the accumulated result by thresholdsegmentation method. Then we can get the leaf area.By analyzing the color characteristics of the leaf, this thesis proposed an algorithmfor leaf occlusion detection based on leaf segmentation and color features. Thealgorithm extracted the color and the leaf area information as the image features. Thenthe classification model was built by support vector machine. The experimental resultsshow that the detection accuracy of this method can reach up to83%.By analyzing the texture of the leaf, an algorithm based on texture and colorfeature fusion was proposed. The algorithm extracted the texture features using UniformLocal Binary Pattern, and fusion it with color features. By feature fusion the robustnesswas improved. The experiments show that the detection accuracy of this method canreach up to85%.