Title: Hybridisation of feed forward neural network and self-adaptive PSO with diverse of features for anomaly detection
Authors: A.R. Revathi; Dhananjay Kumar
Addresses: Department of Information Technology, SRM-Valliammai Engineering College, SRM Nagar, Kattankulathur 603203, Kancheepuram Dt, Tamil Nadu, India ' Department of Information Technology, Anna University, MIT, Chennai, Tamil Nadu, India
Abstract: The objective of this paper is to activity recognition using hybridisation of self-adaptive learning particle swarm optimisation algorithm with feed forward neural network (SLPSO-FFNN). Basically, the system consists of four phases namely, Background Estimation (BE), Object Segmentation (OS), Feature Extraction (FE), and Activity Recognition (AR). At first, we generate the high quality background using BE phase. Then OS model is used to extract the object from the videos and then object tracking process is used to track the object through the overlapping detection scheme. From the tracked objects, the FE module extracts some useful features. Finally, SLPSO-FFNN based approach is used to detect the anomaly present in the videos. In experimentation, we used two types of dataset UCSD pedestrian dataset (Ped_1, Ped_2) and CDNET dataset. The experimental result shows our proposed anomaly detection system obtains the minimum Equal Error Rate (EER) of 0.08% and achieves the maximum accuracy of 97.64%.
Keywords: anomaly detection; EER; SLPSO-FNN; tracked object; object segmentation; background.
International Journal of Biomedical Engineering and Technology, 2018 Vol.26 No.2, pp.111 - 140
Received: 20 May 2016
Accepted: 07 Jul 2016
Published online: 06 Jan 2018 *