Estimating The Last Seen Frame Time of a Static Removed Object in a Recorded Video Feed using Background Subtraction and Feature Extraction

Abstract

ABSTRACT

This research project aims to automate the process of estimating the last seen frame time of a static object removed from a recorded video feed using background subtraction and feature extraction techniques. The motivation behind the study is to enhance the efficiency of video analytics in investigative processes, reducing the need for time-consuming manual reviews of video footage. The system operates using a binary search-inspired algorithm, where the middle frame of the video is analyzed first, and the search is progressively narrowed down to accurately determine the object's disappearance. The system architecture includes five key components: video acquisition, frame preprocessing, background subtraction, feature extraction, and the calculation of the last seen frame time. ORB (Oriented FAST and Rotated BRIEF) is employed for feature extraction, providing a fast and efficient method to detect key object features. The system was tested in a high-performance environment with an Intel i10 processor (2.6 GHz CPU) and 16GB of RAM to ensure fast and efficient processing. OpenCV was used as the primary computer vision library, alongside supporting libraries such as NumPy, Pillow, and Scikit-image. A custom Tkinter GUI was developed for user interaction, and Python was the programming language of choice. While the system performs well in scenarios with simple, low-noise backgrounds, accurately estimating object disappearance times in both short- and long-duration videos, it faces challenges in more complex environments. The system's performance deteriorates in the presence of complex backgrounds, noise, and shadows, leading to significant discrepancies in the estimated disappearance times. To address these limitations, future enhancements include improving the feature extraction algorithm, incorporating shadow detection and removal methods, and developing advanced noise filtering techniques. By integrating these improvements, the system can become more robust and reliable across diverse video conditions, making it highly useful for security and forensic investigations. The proposed approach offers a practical and efficient solution for automated video analysis, with promising applications in fields requiring precise tracking and event identification.