Topic: Monitoring the number of people in the video

Topic Introduction

The number of people in the surveillance video is widely used in life, the effective grasp of real-time information, for the flow of people control, public space design, pedestrian behavior analysis, accident control and so very important. However, this subject also faced with the surveillance video resolution, crowd occlusion and other issues, the current research field of this topic active, put forward a lot of effective methods, but the dependence of the scene is very high, statistical precision also needs to be improved. This graduation design, will be based on the video analysis, combined with the current known solutions, the use of computer vision technology and pattern recognition of the theoretical basis, to carry out the full experiment and thinking, in-depth research and exploration of the subject, put forward to solve the monitoring video of the number of efficient and feasible method of statistics.

Literature Search Overview:

At present, the implementation methods of the number of monitoring video are divided into three categories:
One, through the detection method. Running a pedestrian detector on a single frame image, or detecting a pre-background split, the number of people to be detected is required; This method relies on the performance of the detector, when the population density is higher, the person's attitude change is larger or the appearance of occlusion and other phenomena, the detection performance is poor, and the time consumption of the operation is larger;
Second, through the method of moving clustering. Through the change between the video and the frame, we cluster the people with relative motion, and then use the correlation algorithm to carry out the population statistics. This method requires that the video frame rate is sufficient to support motion analysis, if the actual monitoring video frame rate is too low, it is not applicable;
Third, through the method of statistical regression. Extracting the lower-level features (or pixel levels) of the pre-background, and mapping them to statistical numbers through appropriate regression methods; The selection of appropriate low-level features and reasonable robust mapping methods has a great influence on the regression results.
Currently commonly used data sets have UCSD pedestrian database and PETS2009 two are indoor data sets such as outdoor library and mall dataset, the commonly used evaluation indexes are mean absolute error, mean variance, mean deviation, and the performance of each classical algorithm is greatly improved, in which the mall DataSet, for example, the best performance is the average absolute error of 2.50, the mean variance is 10.0, the deviation is 0.080 (2015,ICCV);
In addition to the sheer number of people, the current research field is also in the development of more research, such as the number of people only focus on the number, the loss of human location, movement behavior, cluster space distribution and other information, these problems are urgently to be resolved; can be counted and monitored video of my wife's monitoring, human tracking, People's behavior recognition and other information combined to give a more comprehensive and in-depth video analysis.

Work plan

First of all the relevant papers in-depth research work and preparation, including the method and implementation of the paper, the comparison of the merits of the methods, to find the current shortcomings of the methods, and then determine the graduation design mainly for the problem and the scene, put forward their own innovative points and technical points, and related technical preparation, determine the experimental methods and technical implementation, In each database and the results of the previous survey, reflect on the improvement, and constantly improve, the completion of the graduation design thesis writing and actively prepare the mid-term defense and final reply;
Schedule: 2016.1.10~2016.2.20, preliminary research, summary thinking;
2016.2.20~2016.2.28, determine the graduation design implementation framework, technical preparation;
2016.3.1~2016.3.31, realize the design, reflect and improve;
2016.4.1~2016.5.1, the leakage of the vacancy, increase the strength of the experiment, write a revised paper;
During the period of writing and research on the work of predecessors, and seriously treat the reply;

Topic: Monitoring the number of people in the video