Deep learning-based facial recognition in image data and detection of physical violence incidents in terms of forensics

It has been observed that the rapid development of technology has spread to almost all areas of our lives. Looking at the developments in the field of artificial intelligence in recent years, it is seen that there are many studies that provide detection, analysis and semantic integrity above the human success rate. Forensic informatics is one of these fields and strives to integrate artificial intelligence into our lives with technological developments. Examination and analysis of forensic evidence can be shown as the problems experienced in the field of forensic informatics. The analysis to be performed by specialized personnel on the files extracted from forensic evidence has great disadvantages in terms of both time and cost. Especially forensic analysis software has limitations on multimedia data such as images, video and audio. In this thesis, in order to overcome the problems that occur during the forensic evidence analysis phase of the forensic informatics process, it is aimed to identify the person/persons and events in the evidence in order to reveal the material truth. Thus, by making person and event correlation, semantic inferences of the content of the data contained in the forensic evidence will be made. In this thesis, it is aimed to perform person detection, recognition and event detection using YOLOv8, one of the real-time and high-performance object recognition architectures in recent years. When we look at the results obtained, our most successful models can perform face detection at 55fps with 99% success, event detection at 55fps with 90% success and identification of people at a rate of 99.8% depending on the hardware. In addition, an interface was created as a result of the thesis study in order to make the detections easily used by forensic informatics experts and the demo version was presented to the service of users.