REAL TIME MULTI-OBJECT IDENTIFICATION AND TRACKING IN PANORAMIC VIEW

Abstract

Computer vision needs to play its role to lessen the burden of security problems. Image tracking over critical areas helps a lot in this regard. A lot of work has already been done on image tracking but the problem lies with the speed of the processing of images. In our project, we use algorithms that not only provide fast tracking but also a greater field of view. One major consideration and concern is regarding the use of suitable hardware, for which we test both our algorithms on various machines, with different operating systems. The fastest and most accurate results are produced by setting up a pipeline system using socket programming. The method we apply for socket programming is pipeline. Two camera feed is given as input to system1 stitching algorithm is running on it. It stitches the feed of both cameras making a panoramic view and give an output to the system2. The system2 will take system1 output and the third camera feed as input and again stitching algorithms will be running on it stitch the input and make a panoramic view. Also the detection and tracking algorithm will be running on the system2 to track objects mainly humans as we are using Tracking algorithms which only detect human beings, detection algorithm will be running on the panoramic view. This aspect also makes our work unique in comparison to that done in the past. So far, majority of the work done in these field only deals with single object tracking over stitched images, which, again is a new feature added in our project