Hidden object detection for classification of threat

摘要

The automated video surveillance has become important due to the focus from government and users for improving the smart nature of the buildings. A system developed for handling this can be used for prison, airport, banks, etc. Though there are solutions for this they fail in situations of mishaps and objects that are hidden that could become a threat to the environment. In this paper a framework has been built using modified K-means segmentation algorithm to detect hidden objects. The framework operates in two phases: phase 1 – modified K-means segmentation algorithm for segmenting the hidden objects; phase 2 – deep convolutional neural network for classifying the hidden object the algorithm selects searched for the approximately optimal value of K and segments the object. The result of the algorithm is given to deep convolutional neural network for classifying the type of object. The algorithm is tested with manually built dataset using Fluke Tis40 Thermal Imager. The experiments were carried out in batches of 50*50 images and the performance of the approach is presented using top-1 accuracy and mean average precision and they are 0.94 and 0.64, respectively. From the experimental analysis, we infer that the proposed algorithm works with precision 0.88 false discovery rates 0.12.