| Expressways are a critical part of infrastructure,playing an essential role in the national economy and people’s daily lives.As expressways continue to rapidly develop,their total mileage has significantly increased throughout the country.While this has brought convenience to people’s travel,it has also led to an increase in the probability of accidents caused by pedestrians entering expressways.Pedestrian abnormal intrusion poses a significant threat to the safety of both vehicles and pedestrians.As a result,pedestrian abnormal intrusion detection has become an urgent measure to ensure the safety of pedestrians and vehicles.However,traditional pedestrian intrusion detection methods for artificial expressways are inefficient and cannot continuously work to ensure detection accuracy.Additionally,these methods are challenging to apply to various sections of expressways.Machine learning-based pedestrian anomaly intrusion detection algorithms for expressways have poor generalization and flexibility.These methods are also difficult to meet the accuracy requirements of pedestrian anomaly intrusion detection under conditions of obstructions present or pedestrian target scale changes.To address these issues,this paper introduces a target detection algorithm based on deep learning into pedestrian abnormal intrusion detection on expressways.The specific main work is described as follows:This study focuses on developing a pedestrian abnormal intrusion detection method for expressways based on a small number of samples.Traditional manual detection methods are often costly,inaccurate,and limited by a lack of samples.To address these issues,we used a pre-trained YOLOv5 s model to detect pedestrian abnormal intrusion on expressways,even in the absence of data sets and algorithm training.Additionally,we introduced a cascade false alarm filtering module to reduce the false alarm rate of the pre-trained model,and a special personnel classification module to filter out special personnel on expressways.Experimental results demonstrate that the accuracy rate of pedestrian abnormal intrusion detection based on fewer samples on expressways reaches 97.84%,which is 55.93% higher than the original algorithm.Moreover,the false alarm rate is as low as 1.05%,which is 57.09% lower than the original algorithm,and could meet the detection performance requirements for abnormal pedestrian intrusion on expressways in production sites.These results demonstrate the efficacy of our approach,which could help to improve the accuracy and efficiency of pedestrian abnormal intrusion detection on expressways.(2)This study proposes an expressway pedestrian abnormal intrusion detection algorithm,YOLOv5-FPI,based on frequency channel attention to address complex detection scenarios where sufficient data is available and higher performance requirements exist.By introducing a frequency channel attention module,a spatial pyramid pooling fast module,and a weighted bidirectional feature pyramid module into the neck network of the YOLOv5 l algorithm framework,this paper aims to improve the network’s ability to extract features from small target pedestrian samples and difficult-to-detect pedestrian samples,thereby enhancing the detection ability of pedestrian abnormal intrusion.Experimental results show that the YOLOv5-FPI algorithm achieves an accuracy rate of 91.09%,a recall rate of 83.25%,and an average accuracy of 89.52%,which is 1.37% higher than the original YOLOv5 algorithm.These results demonstrate the efficacy of our proposed algorithm in improving the detection performance for pedestrian abnormal intrusion in complex scenarios with sufficient data.(3)To meet the need for observability and operability in pedestrian abnormal intrusion detection on expressways,this paper has developed and implemented a software system that integrates three modules: dataset production,model training,and detection.This system enables efficient and effective detection of pedestrian abnormal intrusion on expressways. |
PDF Full Text Request