Object Size Measurement from CCTV footage using deep learning

Abstract

Closed-circuit television (CCTV) cameras are being extensively installed in the cities, apartments, homes, restaurants, educational institutions, shops and so on producing lots of data every second. Storing, processing and analyzing these video footages effectively is a challenging task. In the course of analysis of these video sequences, measurement of size of the objects in the scene is an interesting and a very challenging task. Though there are systems that can determine the height of a person, they are not computationally cost effective and very few of these works for CCTV footages.There are so many applications to find out the size of an object or a person with higher cost of maintenance and there is no application which finds the height of a person in video captured by CCTV. This project mainly focuses on the detection and calculation of height of a person in the CCTV footage using deep learning algorithm.The main objective is to build the system which determines the measurements of an object or a person in a pre-recorded video or CCTV footage or in image. The methodology adapted in the project is to find the bounding boxes around all the objects found in an input and is to determine the object class it belongs to. The python programming and machine learning techniques like YOLO algorithm, R-CNN is used for the object recognition.The posture of a person changes accordingly as his/her actions.The posture of a person effects the calculation of height as likely in a bending position of knees, the height may look less and determined accordingly. The input dataset is having various numbers of predefined classes in it, so that the detection of the various objects in the frames will be easier. The model detects the objects in images and objects in CCTV video footage with 81% accuracy. The custom dataset has been used to train and test the model.