Page 85 - Kỷ yếu hội thảo khoa học lần thứ 12 - Công nghệ thông tin và Ứng dụng trong các lĩnh vực (CITA 2023)

69

2.2 Experimental Setup

The system implementation process is done automatically through 4 steps: Training
image processing, synthesis, comparison and prediction of results.

Challenges. When identifying images of weapons, there are many factors that can
affect the accuracy and recognizability of models:
+ The distance of the Camera to the object is not fixed, in the process of using the
weapon, the object can move continuously, in case the object is too far away, the image
and video capture the object image very well. small, easy to confuse, difficult to model.

Fig. 4. The long distance makes the identification process difficult

+ In many cases, the light is not good enough, the weapon is used in the dark with
very low brightness, or the distance is long, the subject moves in the frame, making the
image not clear, blurred, noise, affecting the input quality of photos and videos.

a. Good light, blurry image b. Poor light image, long distance

Fig. 5. The image of the Pistol from the surveillance camera is blurred and noisy

+ Weapons are often hidden by hands, by many surrounding objects. In particular,
in many cases, they are often placed in carriers and containers such as pants pockets,
bags, etc. When people use weapons for bad purposes, they often intentionally hide
them and use them within a short period of time. Time quickly, so it is difficult to detect
and give early warning.

ISBN: 978-604-80-8083-9 CITA 2023