Research On Key Algorithms For Identifying Key Violations At Oilfield Construction Sites

With the rapid development of society and economy,the demand for petroleum energy is increasing,and the production volume is gradually increasing.However,the security problem restricts the development of oilfield.At present,the identification of violations mainly depends on manual inspections,which has high work intensity and low efficiency.At the same time,there are a large number of cameras in each well site,with high height and long distance.Therefore,the size of targets is small,the target detection is difficult,and the detection results are unstable.In order to realize the intelligence of oilfield video monitoring,the following researches are carried out:(1)A method for oilfield construction scene recognition based on equipment detection is proposed.The K-means algorithm is used to cluster the equipment sizes to obtain 9 prior box sizes.Then,the adjusted prior box is used to improve the darknet-53 network so that it can quickly and effectively identify oilfield equipment.The construction scene is classified according to the detected equipment,and the accuracy rate is 91.38%.(2)A recognition method for answering the phone based on attitude estimation is proposed.First,the application effects of three methods for obtaining human skeleton key points in oilfield monitoring environment are compared.Then,the key points of the skeleton are obtained using Alpha Pose.According to the combinations of different key points detected,the corresponding pose estimation methods are analyzed.In order to reduce the unstable results,a pedestrian tracking algorithm based on the intersection ratio of the bounding boxes between adjacent frames is implemented.Using oilfield monitoring video for testing,the accuracy rate is 82.04%.(3)A method for labor insurance supplies wearing detection based on attribute recognition is proposed.First,based on the improved Res Net network,the multi-label classification problem of the pedestrian upper garment color,lower garment color and whether to wear a hat is solved.Then,combined with Martket-1501 dataset,PETA dataset and oilfield construction scene pictures,a sample-balanced attribute recognition dataset is constructed for network model training.The experimental results show that the recognition accuracy of each attribute is above 94%.Finally,pedestrian tracking algorithm is added to accumulate recognition results,and information entropy is added to reduce unstable results.The detection accuracy of labor insurance supplies wearing is86.19%.The research and application of oilfield construction scene identification and violation behavior identification methods have fully utilized the real-time and initiative of the monitoring system,and provided strong support for oilfield safety production and intelligent video monitoring.