A novel feature extractor for human action recognition in visual question answering

Silva, Francisco H. dos S; Bezerra, Gabriel Maia; Holanda, Gabriel B.; Souza, João Wellington M.; Rego, Paulo A. L.; Neto, Aloísio Vieira Lira; Albuquerque, Victor Hugo C. de; Filho, Pedro Pedrosa Rebouças

doi:10.1016/j.patrec.2021.04.002

Cited by 8 publications

(6 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Reference [18] focused on contextual abnormal human behavior detection especially in video surveillance applications. Reference [19] proposed a new approach to human action recognition for visual question answering, using a novel feature extractor, multiperson 2D pose estimation, and machine learning technique. Reference [20] proposed a unique attention-based pipeline for human action recognition, utilizing both the spatial and the temporal features from a sequence of frames.…”

Section: Security-related Classification and Analyticsmentioning

confidence: 99%

An Intelligent Security Classification Model of Driver’s Driving Behavior Based on V2X in IoT Networks

Dai

Zhong

et al. 2022

Security and Communication Networks

View full text Add to dashboard Cite

Traffic accidents occur frequently in Internet of Things (IoT) safety system. Traffic accidents are largely caused by drivers’ unsafe driving behaviors in the process of driving. Aiming at the problem of low safety of real-time warning in driving, this paper proposes a model to detect driver behavior. Firstly, according to the driver target detection for positioning, combined with the Pose Estimation to identify the driver in the process of driving a variety of driving behaviors, at the same time, a rating model is built to score drivers’ driving behaviors. Then, by integrating the driver behavior model and evaluation rules, the system can give timely and active warning when the driver makes unsafe behavior in the process of driving. Finally, in the V2X scenario, feedback and presentation are given to users in the form of points. The experimental results show that, in the scenario of Internet of vehicles, the driving behavior rating model can well analyze and evaluate drivers’ driving behaviors, so that drivers can more accurately understand their abnormal driving behaviors and driving scores, which plays a significant role in IoT safety management.

show abstract

Section: Security-related Classification and Analyticsmentioning

confidence: 99%

An Intelligent Security Classification Model of Driver’s Driving Behavior Based on V2X in IoT Networks

Dai

Zhong

et al. 2022

Security and Communication Networks

View full text Add to dashboard Cite

show abstract

“…Holanda et al [3] proposed a rapid human action recognition method using visual question-answering techniques, processing a 2D image frame in a mere 190 milliseconds. Their approach involved an ensemble of classifiers, including Bayesian, Multinomial, Complement, and Gaussian classifiers.…”

Section: Vision-based Harmentioning

confidence: 99%

“…Most State-of-the-Art boxing punch classification methods use IMU sensor data to classify the punches. Some studies use visual data [3,5,6,7,8,9] instead, but either need costly depth measuring sensors or could be faster due to naive trajectory feature extraction. Also, those models use heavy Deep Learning classifiers, which require vast amounts of data, extensive for portable devices.…”

Section: Introductionmentioning

confidence: 99%

Efficient boxing punch classification: fine-grained skeleton-based recognition made light

Baghel,

Rithihas,

et al. 2024

Fifth International Conference on Computer Vision and Computational Intelligence (CVCI 2024)

View full text Add to dashboard Cite

Sports analytics is a field of study that utilizes camera and sensor data to monitor the athlete's performance and health to optimize the player's strategy and increase the success rate. Coaches rely on analytics to scout opponents and optimize play calls in gameplay. With the advancement in artificial intelligence, accessible and in-depth data collection has been enabled. The well-grounded technique for performance evaluation in sports analytics is Human Pose Estimation (HPE). Our focus is on real-time action recognition in combat sports like boxing. Existing state-of-the-art deep learning models are heavily parameterized, so can't be used in real-time in any low-end devices. Apart from this, fine-grained classification in highly dynamic activities in sports are typically performed using sensors only. Our proposed Machine Learning based pipeline provides real-time fine-grained solution for 14 boxing punch types of classification using RGB video only. Our approach includes the implementation of three novel and generalized motion dynamics features that encode velocity as well as acceleration of the pose sequences., 1) Unified-Axis Angular Encoding (UAE), 2) 2D Motion Dynamics Descriptors (2DMDD), 3) Fifth-order Angular Encoding (FAE). We employed classical machine learning algorithms I.e., Support Vector Machine (SVM), Random Forest (RF), and K Nearest Neighbours (KNN) to make a lightweight model and test it on YouTube videos. The average accuracies of pipeline using the proposed features are found to be 55%, 92% and 84% for UAE, 2DMDD, and FAE respectively. Using KNN, we have achieved 99% accuracy on 10-fold crossvalidation by using FAE features.

show abstract

“…Run the timing model identification. This paper uses a fixed feature extraction network and an action classification loss function to explore the performance of four common time series modeling methods on this task, including temporal convolution, temporal clustering, temporal attention mechanism, recurrent neural network and its variant models, etc [13][14]. Through research and analysis of 11 time series models, fair and comprehensive performance comparison analysis data is provided for research in this field.…”

Section: Related Workmentioning

confidence: 99%

Visualization system of martial arts training action based on artificial intelligence algorithm

Hui

2023

Preprint

View full text Add to dashboard Cite

At present, there are many educational methods in the field of Chinese martial arts education, and various types of teaching institutions such as private martial arts schools, national sports institutes, and general institutions of higher learning have been formed. However, the proportion of such groups who choose to receive professional martial arts education is very small compared to the population development base of China's entire higher martial arts education. The main reason is that martial arts training activities must be standardized and professional, and it is difficult to practice effectively in a home environment. Therefore, this paper develops a visualization system of martial arts training movements based on artificial intelligence system. Firstly, the theoretical basis of skeletal point extraction feature and posture quaternion in martial arts, the visualization of martial arts movements based on expression quaternions and the classification model of martial arts movements based on LSTM are introduced. For the human action visualization part, we use the Ogre rendering engine to display human actions and provide a data interface to control the movement of the virtual 3D human model. Then the paper introduced the visualization system design of martial arts training movements, including: system architecture design, system function analysis, visualization module design. Finally, the system is disassembled through the test environment and evaluation indicators, action recognition effect analysis, and system visualization impact analysis, and the application effect of the system is introduced in detail. I hope it can help more people inherit and develop traditional sports and carry forward the spirit of martial arts. so that Wushu, a national traditional activity, can be better spread and popularized. This paper studies the artificial intelligence algorithm and applies it to the design process of martial arts training action visualization system, thus promoting the development of martial arts teaching.

show abstract

A novel feature extractor for human action recognition in visual question answering

Cited by 8 publications

References 15 publications

An Intelligent Security Classification Model of Driver’s Driving Behavior Based on V2X in IoT Networks

An Intelligent Security Classification Model of Driver’s Driving Behavior Based on V2X in IoT Networks

Efficient boxing punch classification: fine-grained skeleton-based recognition made light

Visualization system of martial arts training action based on artificial intelligence algorithm

Contact Info

Product

Resources

About