Detection and layout rationality evaluation of blackboard areas in teaching videos by integrating Faster R-CNN and 3D convolutional networks

To achieve precise identification of chalkboard regions and quantitative evaluation of layout rationality in instructional videos, this study proposes a dual-branch model integrating Faster R-CNN with 3D convolutional networks, establishing a collaborative chalkboard analysis framework for spatial detection and temporal modeling. By incorporating an enhanced region proposal network and temporal attention mechanism, the model effectively captures structural features and semantic continuity in both spatial arrangement and writing dynamics of chalkboard content. Combining a multidimensional rationality evaluation metric system, the detection results are comprehensively analyzed from aspects such as spatial utilization, logical structure, and information density. A shallow neural network is constructed to achieve score prediction and rationality classification, utilizing expert annotation data for supervised learning. Results demonstrate that this method outperforms comparison models across multiple evaluation metrics, validating the effectiveness of spatio-temporal feature fusion strategies in complex teaching scenarios.