Extended Evaluation of XZ-Shape Histogram for Human-Object Interaction Activity Recognition based on Kinect-like Depth Image

In this paper, we extend our previous work in investigating the performance of XZ-shape histogram for recognizing human performing activities of daily living (ADLs) which focuses on human-object interaction activities based on Kinect-like depth image. The feasibility of XZ-shape histogram as well as general 3D shape descriptors namely; 1) shape distribution, 2) shape histogram, 3) global spin image and 4) local spin image, in recognizing human-object interaction was tested using RGBD-HOI dataset. Moreover, the proposed evaluation framework was formulated to infer the descriptors’ performance. It was found that, the XZ-shape histogram outperformed other general 3D shape descriptors that compares the performance inferred by the area under receiver operating characteristic curve (AUC-ROC). The results of this study not only demonstrate the implementation of 3D shape descriptor in the dynamic of human activity recognition but also challenge the other general 3D shape descriptor in terms of providing low dimension descriptor that capable in improving the discrimination power of human-object interaction activity recognition. Key-Words: human-object interaction; activities of daily living (ADLs); RGBD image; shape distribution; spin image; shape histogram.