Research On Campus Violence Recognition Algorithm Based On Motion And Speech Features

In recent years,with the growth of media's report,the phenomenon of violence in school gradually attracts people's attention.With the development of the Internet,students who are still in school can get access to various types of information from the society,many of which are filled with violent and vulgar messages.Due to curiosity,some students imitate the behaviors obtained from the Internet,making the phenomenon of violence unprecedented serious on campus.Adolescents need to be given right guidance on their growth stages.However,teachers and their parents cannot always pay attention to their states,and victims are often so frightened or shy that they cannot make teachers and parents understand their situation.Therefore in the early stages,violence cannot be stifled in the bud,which seriously affects the physical and mental health of the victims.Taking the popular use of smart phones among students into consideration,this paper puts forward a school bullying detecting scheme which firstly detects students' activity data collection by mobile phone accelerator and gyroscope,and then judge the situation of violence by pattern recognition technology in a timely way.When bullying occurs,this system will identify the results timely and can inform bullying occurrence to the parents and teachers,so as to intelligently monitor students with no intermittent throughout the day.The scheme would protect the physical and mental health of students,and promote the harmonious development of the campus environment.Different from daily behavior recognition based on motion sensors,the violent activities on campus are chaotic,and it is difficult to describe them in a specific form.In order to construct the campus violence activity model and improve the recognition accuracies of the violent activities,the contents of this paper are given as follows:Firstly,considering the characteristics of violent activities and daily activities,the frequency domain motion features as well as the time domain ones are extracted and the activities categories are identified by BP neural network.For the complex activities,the algorithm of decomposing complex activities into basic ones is proposed to improve recognition accuracy.The precision of violent activities recognition is increased from 85.29% to 88.35%,and the recall is increased from 75.77% to 84.63%.Then,an algorithm is proposed to combine the motion features and the speech features to realize the violence recognition.The simulation results show that the algorithm can effectively improve the precision of violent activities recognition to 90.35% and the recall to 90.49%.Finally,considering future application,an algorithm is proposed to achieve the sleep of the system by finding the transition points of the transitionary activities,which effectively saves the hardware energy consumption caused by the classification algorithm.On the basis of keeping the recognition accuracies of the system unchanged,the dimention of the motion and speech joint features is effectively reduced,which improves the computational efficiency of the classifier,and reduces time complexity of the most activities' recognition algorithm to 50%.