Smart Video Surveillance System
- 0 Collaborators
Utilized a deep object detection network (YOLO) to capture an object's movements in the current camera frame which then served as evidence to a Partially Observable Markov Decision model for visual servoing. ...learn more
Project status: Concept
Groups
Student Developers for AI
Intel Technologies
Intel CPU
Overview / Usage
An important problem at the center of artificial intelligence is the design of suitable control policies for intelligent systems. Video surveillance systems help to enhance security and safety of any place where they are installed and thus are being increasingly used in public places such as banks, airports and shopping malls as well as in confidential and critical settings like government and military premises. The increasing availability and lower cost of higher quality cameras lead to multiple camera installations for monitoring tar-
get areas. In this paper, I present a smart video surveillance system for determining the optimal course of actions that can be taken by a camera for monitoring the movement of an intruder in a real time setting. The proposed intelligent system guides the camera by estimating the direction in which the intruder is currently moving, as well as the discretized intruder’s location corresponding to different frames in the video. I used tiny YOLO for capturing intruders location and direction in the video and a Partially Observable Markov Decision Process (POMDP) formulation to solve the decision-making problem of tracking an intruder based on noisy estimates of the object location and object direction. I used a POMDP solver to compute a policy that balances the cost of directing the camera and tracking the person of interest with the benefit of the camera to continuously monitor a target area for possible intrusion detection.
Methodology / Approach
I used tiny YOLO for capturing intruders location and direction in the video and a Partially Observable Markov Decision Process (POMDP) formulation to solve the decision-making problem of tracking an intruder based on noisy estimates of the object location and object direction. I used a POMDP solver to compute a policy that balances the cost of directing the camera and tracking the person of interest with the benefit of the camera to continuously monitor a target area for possible intrusion detection.
Technologies Used
Technology Stack: Python, Java, OpenCV, Convolution Neural Network
Repository
https://github.com/chiravdave/Projects-Papers/blob/master/Smart_Video_Surveillance.pdf