Search

The growing incidence of ASD has made the development of early diagnostic techniques increasingly crucial. The paper provides a systematic review of video-based motion analysis and deep learning (DL) techniques for early diagnosis of ASD. It evaluates related articles published between 2018 and June 2024. We categorize the leading research according to the body parts involved: head pose, hand and arm pose, and trunk pose. And conduct an in-depth analysis of the procedures, feature extraction techniques, and classification models used in these studies. The paper delves into the general processes and methods proposed in the field, offering a detailed analysis. It presents a timeline of the evolution of ASD diagnostic methods and summarizes publicly available datasets, categorized into behavioral and multimodal datasets. The paper also reports the successes and challenges of applying AI in motion analysis for ASD detection. It provides a complete and systematic overview for researchers in the field of ASD.

Integrating computer vision and machine learning with observation-based ergonomic posture risk assessment methods represents a significant advancement in occupational health and safety, which is essential in reducing work-related musculoskeletal disorders. This collaboration has enhanced the efficiency and scope of ergonomic posture risk assessments. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses guideline was utilized to select relevant papers in this review. A total of 583 research articles were retrieved from the Web of Science and IEEE Xplore databases, and 30 articles meeting the screening criteria were selected for detailed analysis. The results were organized into three distinct stages: Data Preparation, Pose Estimation, and Risk Assessment. During the Data Preparation Stage, the data acquisition devices, datasets, experimental conditions, and observation-based ergonomic posture risk assessment tools utilized in each article were discussed. Subsequently, in the Pose Estimation Stage, the human pose estimation techniques employed and their ability to detect wrist details were outlined. Finally, the risk assessment methods were presented, encompassing Directly Assess Risk Levels, Predict Risk Levels Using Traditional Machine Learning Algorithms, and Advanced Risk Prediction Using Deep Learning Algorithms. We propose several directions for future research. By synthesizing the current literature and identifying critical areas for future development, this review aims to provide a comprehensive overview of the intersection between computer vision, machine learning, and observational ergonomic posture risk assessment, with implications for improving occupational health and safety.

With the growing popularity of rowing machine exercises, it is vital to create effective risk-avoidance solutions for injuries caused by improper rowing techniques in non-instructional circumstances. We built a two-stage system. In step I, each frame of the sequence image is then fed into the OpenPose module so that the coordinates of the key points can be extracted. The trajectories of the knees and elbows are crucial indicators for assessing the conformity of a rower’s actions. We subsequently computed the angle situation utilizing the knee and elbow angle data. In phase II, the one-dimensional angle sequence combination is inputted into a one-dimensional Convolutional Neural Network (1D CNN) to recognize whether the movement is standard. We incorporated a CoT attention module to enhance the classification network’s feature extraction stage. This addition results in highly condensed and information-rich feature representations. In addition, we evaluated four distinct attention mechanism methods for their performance on this test. We collected 665 correct and 490 erroneous rowing action sequences in all. Each sequence has 200 angle data. 75% of each dataset is randomly assigned for training purposes. The remaining 25% is designated as test data. The results of the trial showed an accuracy of 96.65%. It was demonstrated that it is feasible to use a real-time AI method to detect improper actions on a rowing machine through monitoring.

If toddlers are not promptly checked and rescued after falling from relatively high locations at homes, they are at risk of severe health complications. We present a toddler target extraction method and real-time falling alarm. The procedure is executed in two stages: In stage I, a GELAN-integrated YOLOv8 model is used to extract the body features. Based on this, a head capture technique is developed to obtain the head features. In stage II, the “safe zone” is calculated through Generalized Hough Transform (GHT). The spatial location is compared to the preceding stage’s two centers of mass points, K for the toddler’s body and H for the head. Position status detection is performed on the extracted data. We gathered 230 RGB-captured daily videos of toddlers aged 13 to 30 months playing and experiencing upside-down falls. We split 500 video clips (×30 FPS) from 200 videos into 8:2 training and validation sets. A test set of 100 clips (×30 FPS) was cut from another 30 videos. The experimental results suggested that the framework has higher precision and recall in detection, as well as improved mean average precision and F1 scores compared to YOLOv3, v5, v6, and v8. It meets the standard FPS requirement for surveillance cameras and has an accuracy of 96.33 percent.