Concentration Evaluation by Lightweight Neural Network Model with Multi-Face Landmark and Multi-Object Motion Detection

Multi-target concentration evaluation in classroom environments presents substantial challenges for pattern recognition due to concurrent facial variations, diverse distracting behaviors, and real-time processing requirements. Existing approaches primarily rely on single-subject facial analysis and rarely integrate facial landmark information with object-level motion cues in multi-target scenarios. To address these limitations, this paper proposes a Concentration Evaluation by Lightweight Neural Network Model with Multi-Face Landmark and Multi-Object Motion Detection (CEMLN) that jointly models multi-face landmark regression and motion-based distraction recognition for multi-target concentration analysis. The proposed approach integrates facial geometric representations with behavior-related motion patterns, enabling unified detection of distraction, fatigue, and abnormal behaviors under computationally constrained settings. The lightweight network architecture and joint optimization strategy are introduced to achieve an effective balance between recognition accuracy and inference efficiency. Experimental results demonstrate that the proposed method achieves up to 90% concentration evaluation accuracy while maintaining real-time performance at 111 frames per second, validating its effectiveness for practical multi-target visual attention modeling.