Silhouette-Based On-Site Human Action Recognition in Single-View Video
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
On-site worker observation is a fundamental task for a wide spectrum of construction applications such as safety behavior monitoring and productivity analysis. Vision-based action recognition techniques have been proposed to complement the time-consuming and labor-intensive tasks involved in manual observation. In construction, however, previous studies have mainly utilized an RGB-D sensor (e.g., Microsoft Kinect), the operating conditions of which (e.g., active ranges from 80 cm to 4 m, sensitivity to sun light) may hinder the application to actual construction jobsites. To address these issues, we propose a silhouette-based human action recognition method using a single video camera that has less operational constraint. In this framework, the human worker is localized and tracked throughout the monocular video, based on both spatial (i.e., contour of worker) and temporal changes (i.e., moving direction and speed over consecutive frames). Then human action models are learned with temporally adjacent frames and utilized to recognize similar actions in testing video by computing the similarity between the learned action model and newly computed model in a testing dataset. For performance evaluation, we carried out lab experiments, in which a video camera was installed 5–10 m from multiple human subjects. Results indicate that the proposed framework performs well (i.e., an accuracy of 90.68%) to capture predefined poses (e.g., walking, lifting, crawling) in image sequences. This study thus explores an automated means for worker monitoring which potentially helps understand and measure human motions without significant human effort.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it