The Line Monitoring Algorithm (LMA) has demonstrated remarkable efficiency in motion detection for specific line areas in driving test supervision. However, its limitation lies in detecting only whether a line is crossed without providing contextual information about what crossed the line or which vehicle part was involved. This paper proposes a novel hybrid framework that integrates the computational efficiency of LMA with the contextual understanding of deep learning-based object detection. Our approach uses LMA as a high-efficiency trigger mechanism to activate a lightweight YOLO-based object detector only when line crossing events occur. This architecture maintains the real-time performance of LMA while gaining rich contextual awareness. Experimental results on driving test videos show that our hybrid system achieves 96.8% accuracy in line crossing detection while reducing false positives by 73% compared to standalone LMA, with only an 8% decrease in frame processing rate. The system successfully identifies specific vehicle components (tires, body) involved in boundary violations, providing comprehensive evaluation metrics for driving test assessment.
The Line Monitoring Algorithm (LMA) has demonstrated remarkable efficiency in motion detection for specific line areas in driving test supervision. However, its limitation lies in detecting only whether a line is crossed without providing contextual information about what crossed the line or which vehicle part was involved. This paper proposes a novel hybrid framework that integrates the computational efficiency of LMA with the contextual understanding of deep learning-based object detection. Our approach uses LMA as a high-efficiency trigger mechanism to activate a lightweight YOLO-based object detector only when line crossing events occur. This architecture maintains the real-time performance of LMA while gaining rich contextual awareness. Experimental results on driving test videos show that our hybrid system achieves 96.8% accuracy in line crossing detection while reducing false positives by 73% compared to standalone LMA, with only an 8% decrease in frame processing rate. The system successfully identifies specific vehicle components (tires, body) involved in boundary violations, providing comprehensive evaluation metrics for driving test assessment.
