Listening Rate Evaluation System:A Deep Learning Approach

With the development of higher education in our country and the increasing number of college enrollment,the quality of graduates,however,decreases year by year,which leads to the quality of teaching becoming the focus of the society.To this end,universities and collages propose a variety of teaching quality evaluation programs,in which the student quality of lectures undoubtedly become one of the important measurement.However,the existing quality of lecture assessments are basically to rely on man-made,which reduce the efficiency of the assessment and cause a waste of human resources.Therefore,in this paper we propose a learning rate evaluation system based on face detection and crowd counting in order to provide an effective quality assessment solution.Because of the clutter background of the classroom and the problems of the existing face detectors,a grouped facial part based face detector(GFP-FD)is proposed in this paper,which is different from all of the current state-of-art face detectors and take the relatively fixed spatial relation of facial parts into consideration,which is of great help to detect the faces that are under severe occlusion and significantly improves the accuracy of face detection.According to the proposed face detector and face classification,we obtain the initial listening rate of the classroom.To improve the accuracy of listening rate evaluation further,a face detection and crowd counting based approach(GFP-CC)is proposed in this paper,which calibrates the results obtained from GFP-FD and overcomes the limitations of the evaluation based on GFP-FD.As for the experiment result,the proposed face detector(GFP-FD)achieves high recall rate or accuracy on the standard dataset including FDDB,AFW and PASCAL,which outperforms the most state-of-art face detectors.Especially on FDDB,GFP-FD achieves a high recall of91.56% and on PASCAL,GFP-FD achieves a high average precision of 92.48%.When referring to the listening rate evaluation approach(GFP-CC),it owns a high accuracy on the test dataset according to the calibration of crowd counting,which achieves a low mean square error of 6.89% and outperforms the existing approaches.Therefore,the proposed approach(GFP-CC)solves the existing problems of the evaluation of listening rate.