Results (
Vietnamese) 2:
[Copy]Copied!
recognizing its pose in 3D which has been a challenge for video-based methods. Thus, recent computer vision studies focused on using depth maps for conducting activity recognition [38-42], Similar to that in the Architecture/Engineering/Construction (AEC) community, detecting and tracking skeleton and body pose estimation have quickly enabled re¬search on vision-based methods for monitoring worker safety, health monitoring, and activity analysis [8,43-45,71], Weerasinghe et al. [71 ] proposed a method of detecting and tracking workers in 3D using RGB- D sequences. Han et al. [46] studied consequent motion-analysis tech¬niques to detect the unsafe activities of workers by transforming the mo¬tion data onto a three-dimensional space and learn classifiers to detect unsafe activities. Soumitry and Teizer [44] proposed a rule-based classification of worker activities of as ergonomic or non-ergonomic, beneficial for worker training, education, safety, and health. Escorcia et al. [8] pro¬posed a discriminative learning/inference model for classifying activities conducted by single workers self-contained in short videos. As a first step, their method classified single activities per RGB-D sequence con¬taining one worker, however it did not model the variability in duration of single activities nor and the frequency or sequence of their transitions in a long sequence from one activity to another.
Over the past few years, many studies in the computer vision com¬munity have focused on activity recognition from various sequence lengths of RGB-D images (e.g., [38-42]). Sung et al. [39] proposed a method based on detecting and tracking body skeleton and representing them as histogram of gradient (HOG) features for generic human activity recognition from long sequences of RGB-D images. The structure of each activity was learned through hierarchical maximum entropy Markov model and dynamic programming. The RGB-D images used in their experiments were mainly first-person views consisting of simple activities captured under controlled conditions; for example brushing teeth, drinking water, cooking, and opening a
Being translated, please wait..
