Online and Offline Object Detection in Video For Forensic Analysis


  • Muhammad Nur Alam Roslan Malaysian Institute of Information Technology Universiti Kuala Lumpur


In the era of digitalization, many companies, agencies, and even homes in the world have installed closed-circuit television (CCTV) in their premises as surveillance and security purposes. The footage videos from CCTV have been used as the sources of the evidence in the criminal investigation. In video forensic analysis, the footage of CCTV with the potential subject or object is extracted out from the CCTV recordings for the analysis. Law enforcement agencies depend on forensic video analysis software in their process of evidence extraction. However, the majority of the current software is expensive and complex, but it also needs time to analyze. Most of the proprietary forensic software vendors do not provide open access to their source code Sometimes, it does not provide the feature that the investigator needs to conduct the investigation based on certain cases. Object detection is the main key to digital forensics of evidence from CCTV videos. The process of digital forensics can be difficult and it may require a high level of video analysis of lengthy contents that acquire from CCTV cameras. Every image and video collected from CCTV is counted as evidence to create visual documentation of the crime scene. The main purpose of this paper is to assist the forensic investigation by developed an output for an object detection system that able to speed up the process of the extraction of evidence. At the same time, it able to detect objects without human involvement or any external control. To assist the video-based forensic analysis, a convolutional neural networks-based algorithm object detection known as YOLO: Real-Time Object Detection is proposed that can detect and identify potential objects from CCTV footages. This proposed solution able to extract the detected image with a timestamp in online video (such as CCTV and webcam) and offline video (such as CCTV footage and recording video). These implements include deep learning techniques, Graphics Processing Unit (GPU) computing, and efficient, integrated architecture development both for real-time and post-processing.