Smart surveillance cameras are placed in many places such as bank, hospital, toll gates, airports, etc. To take advantage of the video in real time, a human must monitor the system continuously in order to alert security officers if there is an emergency. Besides, for event detection a person can observe four cameras with good accuracy at a time. Therefore, this requires expensive human resources for real-time video surveillance using current technology. The framework of ATM video surveillance system encompassing various factors, such as image acquisition, background estimation, background subtraction, store, and further process like segmentation, people counting, and tracking are done in cloud environment briefly discussed in this chapter.
TopIntroduction
Pedestrian detection is a vital and important task in several smart video surveillance systems. It offers the essential evidence for the semantic understanding of the video footages for video content analysis. It is a new technology for analyzing the video which includes video analytics, text analytics, and audio analytics. From this video, analytics has more challenging and gives a better understanding of the semantics of the video. Video Analytics utilizes numerical calculations to the screen, break down and oversee huge volumes of video. It carefully investigates video inputs; changing them into clever information which helps in making choices.
Video analytics applications can keep running at the inside (on servers or DVRs at the focal observing station), at the 'edge' (incorporated with cameras) or as a mix of both. The 'edge' arrangements are perfect to find live investigation. Focal continuous preparing can come up short on steam in view of the no. of cameras in the system, preparing power and the system data transmission; while in the 'edge' arrangement, each camera has committed handling. Clients with constrained transmission capacity on their systems can settle on an investigation arrangement at the 'edge', so just data on suspicious episodes gets sent through the system; and thus, doesn't go through system transfer speed.
Some run of the mill utilization of Video Analytics in security and surveillance includes, Security Access Point Monitoring, Intrusion Detection / Perimeter Protection, License Plate Recognition, Object Removal, Camera Tampering, Abandoned Object. Current video analytics solutions do work, however, in a compelled domain is a major limitation of video analytics.
Figure 1. Overview of background subtraction methods
Various approaches have been proposed for object tracking. Modeling the object, the suitability of object representation for tracking, selection of features from images are the key requirements of an object tracking algorithms. The choice is made based on the environment in which tracking is performed and the use for which tracking is performed. The significance of object tracking is realized in various activities like motion-based recognition, automated surveillance, video indexing, traffic monitoring, medical applications, industrial applications etc.
Pedestrian detection can be implemented in video analytics techniques which can extract the people from the real-world environment with cost-effective and accurate. It has unlimited potential outcomes in any application territories. An obvious application in surveillance. The quantity of camera in broad daylight places, for example, railway station, shopping centers, and roads develop every year on account of security. And also used in cars, mobile device, flying drone. The immediate reaction in the reason for an occurrence anyway required the manual perception of the video stream which is as a rule financially infeasible. This implies these cameras are for the most part used to catch confirm material after the episode.
In real time pedestrian detection system comprise the techniques like part-based detection, motion-based detection, holistic detection, detection using multiple cameras, patch-based detection for extracting the people from a video scene. Each technique has their own limitations. Holistic detection method uses the mechanism of template matching with the trained model. The drawback is a new template or environment change not handled by this method. Part based detection method based on a collection of part models application detect the object. But the difficulty is detection parts from a video sequence more challenging. Patch based detection is similar to template match but here template is a small block to store in the codebook. Motion-based detection is used to detect foreground object from a video sequence. To perform this process lot of techniques are their background subtraction is one among them. The limitation is to get better background model to detect foreground object. Detection using multiple cameras has to get a 3D view of an object. So, it easily detects the object. But it required more storage and a high processor for processing the data.