An Improved Visual Detection Model for Oversize Vehicle Intrusion in Transmission Corridors

摘要

Aiming at the accuracy and real-time requirements of detection task of oversize vehicle intrusion in transmission corridors, an improved detection method for construction vehicles is proposed. Base on the Yolov4 model, we first analyze the best combination of data enhancement methods for construction vehicles detection. Then we use the K-means clustering algorithm to modify the preset boundary box size of Yolov4. Finally, refer to Soft-NMS algorithm, we improve the process of model boundary boxes detection. The experimental results shows that the average accuracy of the detection can reach 94.15%, which is 12% higher than the original Yolov4. This method can be applied to the detection of dangerous intrusions for the protection of transmission corridors.