Research On Behavior Recognition Algorithm Of Pedestrian Wearing Earphones For Safety Assisted Driving

Pedestrians are one of the most vulnerable traffic groups.Pedestrian traffic accidents occurred so frequently that pedestrian safety must be insured.In the aspect of Safety Driving Assist System(SDAS),the research on pedestrians mostly focuses on pedestrian target detection,and the analysis of pedestrian behavior needs to be improved and supplemented urgently.With the popularity of electronic products,it is common for pedestrians to wear earphones.Pedestrians wear earphones and entertain themselves everywhere.It exposes itself to potential collision hazards and also causes interference to other traffic participants.The safety situation is even more severe.To this end,in this paper we take pedestrian who wearing earphone as research object,based on YOLOv3-Max network,researches the pedestrian recognition in the street intersection scene;and based on pedestrian recognition,applying Adaboost algorithm and SVM algorithm to identify pedestrian wearing earphone behavior and issue warnings to drivers timely,improve related research on safety-assisted driving.The main research contents of this paper are as follows:1)Construct a Caltech-Mix pedestrian recognition hybrid data set.Considering the Caltech pedestrian data set lacking of diversity and according to the characteristics of large-sized pedestrian in the research scenarios of this paper.Taking 5206 images screened from the Caltech data set and 6640 self-collected pedestrian images as the main body,supplemented by the INRIA data set and the CVC pedestrian data set,construct pedestrian recognition data set which called Caltech-Mix hybrid data set suitable for the scenario in this paper.2)Designed and trained a YOLOv3-Max pedestrian detection model based on the YOLOv3 network.For the identification of pedestrians who crossing the street,improve the anchor box mechanism of YOLOv3 network and optimize the network output.The single target pedestrian recognition network YOLOv3-Max for large-sized pedestrians is designed,which greatly improves the real-time performance of the algorithm on the basis of ensuringaccuracy and robustness.The comparative experimental analysis results show that the YOLOv3-Max network has an accuracy rate of 98.3% for large-scale pedestrian recognition.At the same time,the detection speed has increased by 25.6% compared to YOLOv3 network,reaching 49 FPS.3)Design a discrimination algorithm for pedestrian wearing earphones based on Adaboost algorithm and SVM classifier.On the basis of pedestrian recognition,combined Haar-like feature with Adaboost algorithm in ear location detection,feature extraction is performed on the pedestrian detection frame area and the ear area is segmented using a strong cascade classifier.After narrowing the search domain to the ear area,for the unique texture features of the outer contour of the human ear,LBP is selected as the feature operator to extract the features of the ear and use the SVM classifier to determine whether the pedestrian wears earphones.The recognition accuracy of this part of earphone detection on the 1000 ear test images of the self-built EAR-H data set reached 89.1%.4)Construct a behavior recognition system for pedestrians wearing earphones for safety-assisted driving.On the basis of the above research,using the Windows operating system as the platform,the Python programming language was used to construct a pedestrian wearing earphone detection system for safety-assisted driving.Relying on the py QT design interface,the Open CV image processing library is used for algorithm design and development,and the above algorithms are integrated into the same monitoring system.It has been tested that the system can be used for intelligent identification of whether pedestrians wear earphones in real-time video,in order to promptly warn drivers and ensure pedestrian safety.