Development Of Vision-Based Anomaly Detection System In A Dynamic Environment Using Background Subtraction

Motea , Zakarya Mohammed Nasser Saleh (2016) Development Of Vision-Based Anomaly Detection System In A Dynamic Environment Using Background Subtraction. Project Report. UTeM, Melaka, Malaysia . (Submitted)

Text (24 Pages) Development Of Vision-Based Anomaly Detection System In A Dynamic Environment Using Background Subtraction.pdf - Submitted Version Download (640kB)

Abstract

Recently, surveillance systems have received a great interest from various users throughout the world. Some of these systems have started using background subtraction which is able to observe the activities in progress. So far, some of background subtraction based systems have been able to detect moving objects, however, the small view and the high rate of data still a technical challenge. The previous works done on this field by researchers still face some challenges which degrade the performance parameters; the precision, computing times, adaptability with multimodal backgrounds. Moreover, there are some challenges with applications and environment including the change in illumination, dynamic backgrounds or the blur caused by bad weather conditions and so on. The objective of this project is to evaluate and analyses the performance of Gaussian Mixture Model algorithms (GMM) for subtracting the foreground objects from the background in out-door scenes in different situations including all the challenges encountered in dynamic backgrounds, with bad weather conditions and the change in illumination. Then, the feasibility of GMM for surveillance systems is tested using some Datasets (image sequences and videos) which contain all the conditions and using a webcam. It will contribute to the development of vision based anomaly detection systems for monitoring and providing content-based images and videos recording of object’s motion in a dynamic environment. The proposed method in this project is to use GMM which is suitable for multimodal backgrounds. The procedure to detect and then track the anomaly objects is done using C++ and OpenCV libraries as it contains various programming functions used for image processing. It does process sequences of images, videos and real-time videos streaming from a camera. The result shows that, GMM algorithms are able to detect and track anomaly objects in a dynamic environment. The feasibility of GMM for surveillance systems is proved. The precision of algorithms in most of the conditions tested was more than 90% and in the worst condition during snowy weather it wasn’t less than 50%.

Actions (login required)

View Item