Research On Recognition Of Unsafe Behavior Of Crew Keep Watching Based On Deep Learning

Object detection algorithms and human key point detection algorithms are used in the field of behavior recognition.Many scholars have also research on unsafe behavior recognition in various scenarios.However,there are few studies on the unsafe behaviors recognition of crews watchkeeping and some problems such as false detection and poor realtime performance.Therefore,based on computer vision and deep learning,this thesis studies methods for identifying unsafe behaviors of crews in a watchkeeping environment.The main work are as follows:(1)The concept about unsafe behavior of crews watchkeeping is proposed.Based on the current status of research,this thesis analyzes the classification standards of unsafe behaviors in the transportation industry and the existing problems in current research,summarizes the concept of unsafe behavior of crews watchkeeping;The importance of attention concentration during watchkeeping is summarized from various navigation environments such as narrow waterways,foggy navigation,and fishing areas.A behavior set is formed.(2)A lightweight small object detection model is proposed.Ordinary convolution is replaced to reduce the parameters and computational complexity of the model;The EIOU loss function is introduced to refine the position loss.Use the K-means++ algorithm to meet the size of the research goal.Results show that the improved object detection model improves the detection accuracy of small objects,reduces the computational workload,and accelerates the detection speed of the model.(3)An unsafe behavior recognition algorithm is proposed.Using Blaze Pose human key algorithm.Elbow bending behavior rules and snooze rules are introduced as supplementary conditions to recognition unsafe behaviors during watchkeeping,and the relationship between the object and the human key point is used as the judgment basis for the occurrence of unsafe behaviors during watchkeeping.research is conducted on drowsiness behavior,identifying crews drowsiness behavior through facial key point detection and drowsiness rules.