A morphological approach to detect human in video

Abstract

Human detection in video is a challenging task due to complex backgrounds, occlusions, variations in lighting conditions and so on. The main objective of this paper is to determine the presence of human in a video scene. It finds usage in determining number of persons which is found to be a useful metric in understanding the interested participants and their interaction with the environment. The foreground is detected using Gaussian mixture model and is processed to remove unwanted noise by applying suitable morphological operations forming a binary image. The dominant blob region is identified using connected component labeling technique and averaging methods are employed between clean foreground mask and binary image. Finally, edge detection is applied to each processed frame and edge details in the segmented blob displays the presence of the human in the scene. The qualitative results of the proposed system show improved detection accuracy avoiding missed and false detections.