Camera-based surveillance systems have been widely used to monitor public places for ensuring security and safety. One of the problems in the surveillance process is identifying human objects on different CCTV cameras, which is referred to as person reidentification (Re-ID). Re-ID is the process of identifying whether the images of human objects, captured from two or more images from CCTV cameras with the different viewpoints, are the same person or not. This paper proposes a method based on visual features of the object image, named as Bag of Visual Feature (BOVF). BOVF works by representing image data as a collection of local features that are used with a feature clustering mechanism. BOVF implementation uses the Density Based Spatial Clustering of Applications with Noise (DBSCAN) clustering methods in the Histogram of Oriented Gradient (HOG) features. The results of this study with 70 image frames from iLIDS-VID dataset obtained the best accuracy at R-20 by 88% using DBSCAN with a processing speed of 1.85 seconds.
Copyrights © 2020