Motion Video Data Research Articles

Skeleton action recognition via graph convolutional network with self-attention module

<abstract><p>Skeleton-based action recognition is an important but challenging task in the study of video understanding and human-computer interaction. However, existing methods suffer from two deficiencies. On the one hand, most methods usually involve manually designed convolution kernel which cannot capture spatial-temporal joint dependencies of complex regions. On the other hand, some methods just use the self-attention mechanism, ignoring its theoretical explanation. In this paper, we proposed a unified spatio-temporal graph convolutional network with a self-attention mechanism (SA-GCN) for low-quality motion video data with fixed viewing angle. SA-GCN can extract features efficiently by learning weights between joint points of different scales. Specifically, the proposed self-attention mechanism is end-to-end with mapping strategy for different nodes, which not only characterizes the multi-scale dependencies of joints, but also integrates the structural features of the graph and an ability of self-learning fusion features. Moreover, the attention mechanism proposed in this paper can be theoretically explained by GCN to some extent, which is usually not considered in most existing models. Extensive experiments on two widely used datasets, NTU-60 RGB+D and NTU-120 RGB+D, demonstrated that SA-GCN significantly outperforms a series of existing mainstream approaches in terms of accuracy.</p></abstract>

Video Classification for Video Service Providers : A Survey

One of the very interesting data modalities is video. From a dimensionality and size perspective, videos are one of the most interesting and intuitive data types which enable fast and easy object recognition and learning. Video classification is an important task for archiving digital contents for various video service providers. Video uploading platforms such as YouTube are collecting enormous datasets, empowering Deep Learning research. Videos being an important source to recognize any activity by the humans, video classification becomes an important and critical job for video service providers. The survey paper studies various deep learning, transfer learning and hybrid model approaches. Video data normally occurs as continuous, analog signals In order for a computer to process this video data, the analog signals must be converted to a non-continuous, digital format. In a digital format, the video data can be stored as a series of bits on a hard disk or in computer memory. A video sequence is displayed as a series of frames. Each frame is a snapshot of a moment in time of the motion-video data, and is very similar to a still image. When the frames are played back in sequence on a display device, a rendering of the original video data is created. In real-time video the playback rate is 30 frames per second. This is the minimum rate necessary for the human eye to successfully blend each video frame together into a continuous, smoothly moving image. A single frame of video data can be quite large in size. A video frame with a resolution of 512 x 482 will contain 246,784 pixels. If each pixel contains 24 bits of color information, the frame will require 740,352 bytes of memory or disk space to store. Assuming there are 30 frames per second for real-time video, a 10-second video sequence would be more than 222 megabytes in size! It is clear there can be no computer video without at least one efficient method of video data compression.

Construction of Virtual Video Scene and Its Visualization During Sports Training

This article studies the actual captured human motion data for human motion synthesis and style transfer, constructs a scene of motion virtual video, and attempts to directly generate human motion style video to establish a sports style transfer model that combines and self-encoding. The original human motion capture data mapped to the motion feature space for style transfer synthesis. The coding network used to map the high-dimensional motion capture data to the low-dimensional feature space, and the motion style transfer constraints established in the feature space, and the human body motion results after the style transfer obtained by decoding. This paper proposes a pixel-level human motion style transfer model based on conditional adversarial networks and uses convolution and convolution to establish two branch coding networks to extract the features of the input style video and content pictures. The decoding network decodes the combined two features and generates a human motion video data frame by frame. The Gram matrix establishes constraints on the encoding and decoding features, controls the movement style of the human body, and finally realizes the visualization of the movement process. The incremental learning method based on the cascade network can improve the high accuracy and achieve the posture measurement frequency of 200Hz. The research results provide a key foundation for improving the immersion sensation of sport visual and tactile interaction simulation.

Real-Time On-Demand Motion Video Change Detection in the Sensor Web Environment

Detecting motion-based video change, such as different types of motion video or applications using different change detection algorithms in a Web system, is difficult. This paper designs and implements architecture for a real-time or near real-time on-demand motion video change detection system using the Sensor Observation Service (SOS) and the Web Processing Service (WPS) in the Sensor Web environment. Real-time or near real-time includes sensors that obtain motion video data, SOS provides motion video data to WPS and WPS processes this data. Three solution methods are introduced: the GetObservation operation of SOS by transaction, dynamical interaction between SOS and WPS, and WPS real-time or near real-time processing. On-demand means that a developer can choose different motion video change detection algorithms under different applications or different conditions. For this purpose, a flexible, standards-based and service-oriented WPS architecture is designed, which consists of three layers: the WPS interface layer, the field interface layer and the implementation layer. To test the proposed approach, a video change detection case of monitoring a road situation is shown, which was a demonstration for Open Geospatial Consortium Web Service phase 7. The results demonstrate that the proposed approach is feasible.

Dynamic real-time scheduling strategies for interactive continuous media servers

In this paper, we propose and study a dynamic approach to schedule real-time requests in a video-on-demand (VOD) server. Providing quality of service in such servers requires uninterrupted and on-time retrieval of motion video data. VOD services and multimedia applications further require access to the storage devices to be shared among multiple concurrent streams. Most of the previous VOD scheduling approaches use limited run-time, 0 information and thus cannot exploit the potential capacity of the system fully. Our approach improves throughput by making use of run-time information to relax admission control. It maintains excellent quality of service under varying playout rates by observing deadlines and by reallocating resources to guarantee continuous service. It also reduces start-up latency by beginning service as soon as it is detected that deadlines of all real-time requests will be met. We establish safe conditions for greedy admission, dynamic control of disk read sizes, fast initial service, and sporadic services. We conduct thorough simulations over a wide range of buffer capacities, load settings, and over varying playout rates to demonstrate the significant improvements in quality of service, throughput and start-up latency of our approach relative to a static approach.

Motion Frame Analysis and Scene Abstraction: Discrimination Ability of Fuzziness Measures

This article addresses a solution to the problem of scene estimation/abstraction of motion video data in the fuzzy set theoretic framework. Using various fuzzy geometrical and information measures as image features, an algorithm is developed to compute the change of information in each of two successive frames to classify scenes/frames. Frame similarity is measured in terms of weighted distance in fuzzy feature space. This categorization process of raw input visual data can be used to establish structure for correlation. The investigation not only attempts to determine the discrimination ability of the fuzziness measures for classifying scenes, but also enhances the capability of nonlinear, frame-accurate access to video data for applications such as video editing and visual document archival/retrieval systems in multimedia environments.