Research On State Detection Of Mobile Targets In Open Scenes And Construction Area

In recent years,with the rapid development of relevant technologies in the field of artificial intelligence,the smart site scheme related to the construction industry is gradually maturing.With the rapid development of new technologies,such as automatic detection of the status of goods and events in the decoration and distribution,the corresponding intelligent detection results of the equipment will also be realized;For the moving target state detection in the open scene and construction area closely related to the smart construction site,the existing methods have the problems of single semantic information of the target state detection results,lag,easy loss of targets and high false alarm rate.Therefore,this paper has conducted a series of research on the defects of existing methods in moving target state detection.The main preparatory work and innovative achievements are as follows:(1)The principle and experimental test of the target detection and target tracking algorithm involved in the proposed method are introduced.In target detection,the map value of the test result of yolov3 algorithm is higher than 0.8,and the detection accuracy of frequent targets in two scenarios is high;In the target tracking link,the pedestrian is taken as an example to verify the tracking accuracy and robustness of deep sort algorithm in the moving target state detection link.(2)A moving target state detection method suitable for open scene is proposed.The method based on target tracking and trajectory prediction is used to detect the moving target state in the open scene and summarize the group targets with similar motion state;Taking pedestrians as an example,the characteristics of interaction between targets are analyzed,and their travel trajectory is mathematically modeled.By analyzing the linear equation of trajectory and the temporal and spatial characteristics of target motion,a target static state,parallel state and conflict state detection system is designed;Taking the crowd as an example,a group target induction algorithm with similar motion state is proposed.The experimental results show that the success rate of parallel state detection is higher than 90.9%,and the success rate of conflict state detection is higher than 91.4%;Compared with the existing methods,it can detect various possible states of moving targets 12-32 frames in advance,and filter out the population containing non-parallel state targets,so as to summarize the population with similar motion states.(3)A moving target state detection method suitable for construction area is proposed.In the extraction of unknown type targets,a foreground target extraction method integrating mog2 and yolov3 is designed.Compared with the traditional method,it can remove the known type targets in the environment without setting a large number of geometric thresholds;In the process of unknown target in static state,a method to prevent foreground target from merging into background is proposed,and a target removal state detection algorithm is designed when it moves again,which reduces the false alarm rate in the process.