Interpretation and semantic indexing of an image content

Abstract

The field of object detection has received a lot of interest from researchers in recent decades. As it is estuary for many interesting applications, in many areas of life such as unmanned aerial vehicles, self-driving cars, robots, and many other applications that require recognition of objects and scenes within images. Due to the large number of crimes and work accidents in the last century, our goal is to propose a method for introducing the concept of self-monitoring in the field of security, for example, for a group of people, each individual has his or her unique appearance and body shape, this is what makes it difficult to recognize, and we find that there is still no acceptable approach that gives an accurate semantic interpretation of images compared to the performance of humans, which made access to classifying the dangers to which a person may be exposed with high accuracy is difficult. We created a new dataset dedicated to work accidents. We used YOLOv3 model to verify the presence of humans in the images, then we classified them into two categories (the category of presence of risk and the other being the absence of risk) using the CNN (vgg16 and inception v3) and KNN algorithms. In general, our methodology will provide additional information for the semantic analysis and interpretation of the content of the images, which is our desired result. The result Obtained it is acceptable and promising, so the proposed method helps save lives, but it needs more data.