click here to download project abstract of security cameras
Abstract:
We provided security cameras project in this paper. Nowadays, there has been a rise in the amount of disruptive and offensive activities
that have been happening. Due to this, security has been given principal significance.
Public places like shopping centers, avenues, banks, etc. are increasingly being
equipped with CCTV to guarantee the security of individuals. Subsequently, this
inconvenience is making a need to computerize this system with high accuracy. Since
constant observation of these surveillance cameras by humans is a near-impossible
task. It requires work forces and their constant attention to judge if the captured
activities are anomalous or suspicious. Hence, this drawback is creating a need to
automate this process with high accuracy. Moreover, there is a need to display which
frame and which parts of the recording contain the uncommon activity which helps
the quicker judgment of that UN ordinary action being unusual or suspicious.
Therefore, to reduce the wastage of time and labor, we are utilizing deep learning
algorithms for Automating Threat Recognition Systems. Its goal is to automatically
identify signs of aggression and violence in real-time, which filters out irregularities
from normal patterns. We intend to utilize different Deep Learning models (CNN and
RNN) to identify and classify levels of high movement in the frame. From there, we
can raise a detection alert for the situation of a threat, indicating the suspicious
activities at an instance of time
INTRODUCTION
Presently, there has been an increase in the number of offensive or disruptive
activities that have been taking place these days. Due to this, security has been given
utmost importance lately. Installation of CCTV for constant monitoring of people and
their interactions is a very common practice in most of the organizations and fields.
For a developed country with a population of millions, every person is captured by a
camera many times a day. A lot of videos are generated and stored for a certain time
duration. Since constant monitoring of these surveillance videos by the authorities to
judge if the events are suspicious or not is nearly an impossible task as it requires a
workforce and their constant attention the security cameras. Hence, we are creating a need to automate this
process with high accuracy. Moreover, there is a need to show in which frame and
which parts of it contain the unusual activity which aids the faster judgment of that
unusual activity being abnormal or suspicious. This will help the concerned
authorities to identify the main cause of the anomalies that occurred, saving time and
labor required in searching the recordings manually. Anomaly Recognition System is
defined as a real-time surveillance program designed to automatically detect and
account for the signs of offensive or disruptive activities immediately. This work
plans to use different Deep Learning models to detect and classify levels of high
movement in the frame. In this work, videos are categorized into segments. From
there, a detection alert is raised in the case of a threat, indicating the suspicious
activities at an instance of time. In this work, the videos are classified into two
categories: Threat (anomalous activities) and Safe (normal activities). Further, we
recognize each of the 12 anomalous activities – Abuse, Burglar, Explosion, Shooting,
Fighting, Shoplifting, Road Accidents, Arson, Robbery, Stealing, Assault, and
Vandalism. These anomalies would provide better security to the individuals. To solve
the above-mentioned problem, deep learning techniques are used which would create
phenomenal results in the detection of the activities and their categorization. Here,
two Different Neural Networks: CNN [3] and RNN [4] have been used. CNN is the
basic neural network that is being used primarily for extracting advanced feature maps
from the available recordings. This extraction of high-level feature maps alleviates the
complexity of the input. To apply the technique of transfer learning, we use
InceptionV3- a pre-trained model. The inceptionV3, pre-trained, is selected by
keeping in view that the modern models used for object recognition consider loads of
parameters and thus take an enormous amount of time to completely train it.
However, the approach of transfer learning would enhance this task by considering
initially the previously learned model for some set of classified inputs e.g., Image-net;
which further can be re-trained based on the new weights assigned to various new
classes. The output of CNN is fed to the RNN as input. RNN has one additional
capability of predicting the next item in a sequence. Therefore, it essentially acts as a
forecasting engine. Providing the sense to the captured sequence of
actions/movements in the recordings is the motivation behind using this neural
network in this work. This network has an LSTM cell in the primary layer, trailed by
some hidden layers with appropriate activation functions, and the output layer will
give the final classification of the video into the 13 groups.
