Human Detection in Infrared Imagery using Gradient and Texture Features and Super-pixel Segmentation

Many human detection algorithms are able to detect humans in various environmental conditions with high accuracy, but they lack the ability to give the exact region of where the human is located (usual detections as a bounding box). The proposed algorithm utilizes a two-stage approach for human detection: gradient and texture features and super-pixel segmentation. The first stage is a high accuracy human detection algorithm that uses gradient information through the Histogram of Oriented Gradients and texture information through the center-symmetric local binary pattern. Various binning strategies help keep the inherent structure embedded in the features, which provide enough information for robust detection of the humans in the scene. The second stage is the SLIC super-pixel segmentation algorithm to find the actual regions of the person that are not background information. The bounding box is assumed to have surrounding background information with foreground information as the human. The second stage characterizes the background information surrounding the human and deletes and super-pixel information that contains background information, which then groups the remaining foreground information into a convex hull representation. The algorithm is shown to create a better representation of the human detection for analysis of scenes as compared to normal detection strategies.