Page 81 - Kỷ yếu hội thảo khoa học lần thứ 12 - Công nghệ thông tin và Ứng dụng trong các lĩnh vực (CITA 2023)
P. 81

65


                        Weapon Detection with YOLO Model Version 5, 7, 8




                                                1
                                                                             2
                                                                                            3
                                Doan Trung Son  and Nguyen Thi Khanh Tram , Vo Thai Anh
                                             1  Phenikaa University, Ha Noi, Vietnam
                              2  University of Engineering and Technology, VietNam National University
                                             3 Can Tho University, Can Tho, Vietnam
                                         son.doantrung@phenikaa-uni.edu.vn




                           Abstract. Today, the crime rate due to the use of hot weapons is very high, early
                           detection  of  possible  violent  situations  is  extremely  important  for  security.
                           One way to prevent these situations is by detecting the presence of dangerous
                           objects in surveillance video. Current surveillance systems still require human
                           supervision, are low-precision and costly. AI (Artificial Intelligence) technology
                           is  being  used  in  almost  all  sectors  of  society.  Recently,  the  Smart  Camera
                           application  that  uses  Artificial  Intelligence  to  solve  complex  problems  is
                           increasingly  popular.  This  paper  focuses  on  improving  the  performance  of
                           Surveillance Cameras to detect pistol in both accuracy and diversity by using
                           YOLO -V5, V7, V8 models.

                           Keywords: Hot Weapon, Pistol Detection, YOLO-V5, YOLO-V7, YOLO-V8.



                     1     Introduction


                     According to statistics, criminals use hot weapons to commit crimes such as fights,
                     riots, terrorism, drug trafficking  etc. in social life, in which the risk of security and
                     safety for people is the most important issue today in social security. Recently, there
                     have been consecutive cases related to "hot" weapons, making people really worried
                     and insecure. From the above situation, it is more urgent than ever to tighten arms
                     management and especially introduce early warning measures for people to prevent
                     them. Automatic image recognition is becoming popular and applied in many fields
                     [10] including detecting dangerous situations based on surveillance camera recordings.
                       This article deals with the analysis of footage from video obtained by surveillance
                     camera  systems.  Some  typical  Smart  Camera  systems  include:  IBM  Camera
                     Surveillance System [8] which automatically tracks and detects moving objects in a
                     surveillance area, KNIGHT System [9] which specifically allows subject tracking from
                     multiple surveillance cameras. However, the cost of these solutions is quite high or does
                     not support the implementation of many specific problems in Vietnam.



                     Related Research.  In [1] Olmos proposes A binocular image fusion approach for
                     minimizing false positives in handpistol detection with deep learning. Object detection
                     models have known important improvements in the recent years. The state-of-the art





                     ISBN: 978-604-80-8083-9                                                  CITA 2023
   76   77   78   79   80   81   82   83   84   85   86