Person Re-Identification

876 Words2 Pages

In the field of video surveillance, person re-identification is a task of matching the identity of a person captured by different cameras in the network at different places and times. The cameras used for surveillance are located at a much higher position than the person so that the conventional method of face recognition is not used for identification of the person. The images of the same person may vary from one camera view to another camera view (view point variation), or due to different lighting conditions (variation in illumination) or due to posture changes. In this paper, we introduce a re-identification method based on sparse representation. We have formulated the person re-identification as a ranking problem in which the proposed …show more content…

The person re-identification task is to match a person from one camera view with the images captured by other non-overlapping cameras. This task is highly challenging since the images of the same person captured at various places and times vary notably. When re-identification is performed manually, it requires a laborious effort but still remains inaccurate. With the increase in the use of video surveillance in public places, the interest in automated re-identification is growing. The conventional way of identifying a person in a crowd is done by face recognition, but this method is not possible as it is very difficult to get the details required for extraction of face features. Alternatively, other visual features like clothing, objects associated can be used for re-identification. But, still the visual appearance features remain very weak due to a number of reasons. Firstly, the cameras used for surveillance are fixed at a distance and the environment to be captured is highly uncontrollable. Secondly, when disjoint cameras are used, it is hard to fix the transit time between the cameras, as it varies from person to person. Thirdly, the features extracted from clothing are not distinct because there is a chance that many people wear clothes of the same color. Other important reasons include the variations in lighting conditions, view angles, occlusion, and background clutter.In the person …show more content…

The color features commonly used are the histograms of different color spaces like RGB, HSV, YCbCr and Log Chromaticity. The uses of color features reduce the computational cost. Color features are robust to variations in resolution and perspective [1]. Earlier works used single color space for feature extractions. Cheng, et al. [2] used histogram of RGB color space as the features, whereas Farenzena,et al. [3] used HSV color space. Several works were done to find more discriminative features which use more than one color space along with other features like texture, Histogram oriented gradients (HOG). Li, et al. [4] used the histogram of oriented gradients and color histogram in HSV space as feature representation. Syed Fahad Tahir [5] used features that are extracted only from the upper body using RGB, YCbCr and HSV color spaces for dimension reduction. Liu [6] used Joint HSV histogram as the feature descriptor. Tao,et al.[7] concatenated Local binary pattern (LBP)descriptor, HSV histogram, and RGB histogram were extracted from the overlapping blocks. LBP is a texture feature. Wang, et al. [8] used RGB, HSV histograms concatenated with LBP descriptors. Le An, et al. [9] extracted color features quantized mean values of different channels in the HSV and Lab color space, the semantic color names and the texture feature, Local Binary Patterns (LBP), Gabor

Open Document