Abstract In this paper, a depth camera-based novel approach for human activity recognition is presented using robust depth silhouettes context features and advanced Hidden Markov Models (HMMs). During HAR framework, at first, depth maps are processed to identify human silhouettes from noisy background by considering frame differentiation constraints of human body motion and compute depth silhouette area for each activity to track human movements... In this paper, a depth camera-based novel approach for human activity recognition is presented using robust depth silhouettes context features and advanced Hidden Markov Models (HMMs). During HAR framework, at first, depth maps are processed to identify human silhouettes from noisy background by considering frame differentiation constraints of human body motion and compute depth silhouette area for each activity to track human movements in a scene. From the depth silhouettes context features, temporal frames information are computed for intensity differentiation measurements, depth history features are used to store gradient orientation change in overall activity sequence and motion difference features are extracted for regional motion identification. Then, these features are processed by Principal component analysis for dimension reduction and kmean clustering for code generation to make better activity representation. Finally, we proposed a new way to model, train and recognize different activities using advanced HMM. Each activity has been chosen with the highest likelihood value. Experimental results show superior recognition rate, resulting up to the mean recognition of 57.69% over the state of the art methods for fifteen daily routine activities using IM-Daily Depth Activity dataset. In addition, MSRAction3D dataset also showed some promising results. Read More Read Less

Abstract Image segmentation is a basic but important preprocessing to image recognition in computer vision applications. In this paper, we propose a pixel-based k-means (PKM) clustering to generate superpixels, which comprise many pixels with similar colors and neighbor positions. In contrast with conventional center-based clustering, the PKM method traces several nearer clustering centers for a pixel in advance, and then the pixel find the... Image segmentation is a basic but important preprocessing to image recognition in computer vision applications. In this paper, we propose a pixel-based k-means (PKM) clustering to generate superpixels, which comprise many pixels with similar colors and neighbor positions. In contrast with conventional center-based clustering, the PKM method traces several nearer clustering centers for a pixel in advance, and then the pixel find the highest similar colors as its clustering center. Besides, we adopt the regional clustering of the SLIC (Simple Linear Iterative Clustering) in the PKM method to improve the performance of image segmentations. The MSRC dataset is used to quantitatively compare the PKM with the SLIC performances, such as under-segmentation errors, boundary recall, detection precision, and computation efficiency. Read More Read Less

Abstract Image super resolution (SR) based on self-similarity has achieved impressive results by exploiting the local similarity across various scaled images. However, it is time consuming and far from meeting practical applications’ requirements and it is prone to unnatural visual artifacts such as facets and enhanced noise. In this paper, with the powerful general purpose GPU, an accelerated parallel implementation based on NVIDIA CUDA... Image super resolution (SR) based on self-similarity has achieved impressive results by exploiting the local similarity across various scaled images. However, it is time consuming and far from meeting practical applications’ requirements and it is prone to unnatural visual artifacts such as facets and enhanced noise. In this paper, with the powerful general purpose GPU, an accelerated parallel implementation based on NVIDIA CUDA architecture has been proposed. Then denoising is combined and implemented together during the SR processing using the patch matching information of similarity, and it makes the images clearer and cleaner. Also, subpixel accuracy of interpolation and searching is applied to improve the similarity matching process, which make the upscaled images more natural-looking. Experimental results demonstrate that the proposed method achieves an impressive speedup rate and produces improved visual quality images in comparison with other state-of-the-art SR. Read More Read Less