A System for Movement Detecting Congestion

In this paper, we present a system for tracking and provide early information of hazardous locations in huge gatherings. It is based on optic flow estimations and detects sequences of crowd motion that are characteristic for devastating congestions. For optic flow computation, Lucas- Kanade method is employed to determine the optical flow vectors for the gathered video. Segmentation of video sequences is done and optic flow is determined for respective segments. A threshold optic flow is chosen in such a way that the tracking of congested area in video is easily done by comparing it with respective segment’s determined optic flow values. Finally, we present the location of crowd congestion which helps in taking further protective measures to handle unusual events.


1.INTRODUCTIONS
Around the globe extensive gatherings are most frequent and very popular in socio-activities. Currently, typical examples include sporting events, gatherings at airways, railway stations, famous spots, fairs, or recitals. They attract uncontrollable large number of visitors and major precautious actions are always necessary. In spite of everything, the use of current computer vision area such as video surveillance; lethal human stampedes and crowd tragedy still happen. (refer Table 1). Recently, a huge number of people travel together and may meet in a particular location for a specific purpose such activity is termed as mass gatherings. Rally of politics, gathering at fairs, religious occasions, gathering at railways are the events of mass gatherings. Such gatherings pose severe threats for crowds, because huge number of people moving in small areas can promote the spread of dangerous diseases and enhance the threat of injury which eventually results in crowd stampedes. Crowd can rely on factors, such as individual's emotion, elderliness, and consumption of liquor, which influence the severity of harm. Such factors can make crowds to be more violent. Thus in such places several issues should be addressed well for avoiding unexpected situations. Nowadays, many researchers are showing lots of interest in computer vision area. Study of human movement topic in that area has been popular for modern technical decades. The popularity is due to a huge expansion of applications in video vigilance and crowd dynamics. Because the activity is of immense scientific concern, it offers different computational challenges and because of an expeditious upsurge in video vigilance technology deployed in both private and public areas.
Many research reports conclude that during large gatherings, the density of people rapidly increases; reports show that on an average ten people per square meter density has been recorded [1]. Immense density of crowd naturally occurs from varying pattern zs of crowd behavior such as stop-and-go waves or turbulence in crowd [2]. In order to avoid frequent collision in crowd, they have tendency to transit from a steady flow to a stop-and-go manner, as a result individuals move with reduced footsteps distance and finally stops moving at any point of time..This kind of situation where people are packed densely results in crowd turbulence. Intermittent forward pedestrian movement and backward free space propagation are the characteristic of stop-and-go wave patterns. They are caused when the obstructions and densities of pedestrian are uncontallable. During such uncontrollable situation involuntary pedestrian movement will cause rapid motion of other neighboring people. Thus in such scenario individuals may walk falsely and finally collapse on ground. Subsequently, because of lethal pressure of up to 4500 N/m on pedestrian chests and also deprivation in oxygen leads to loss of life [2].
The motivating factor is to implement a system that is capable of identifying and track crowd motion, which is a major interesting research topic in vision analysis, security, crowd dynamics, and visual surveillance fields. The system must be able to track people movement in the subsequent video frames and should provide early information to the concerned authority about the location of overcrowding. The developed system might find real-time application in visual surveillance, crowd analysis, identifying suspicious people activities in crowd. Thus, developed system can be enhanced to address security issue in immense crowding. In this paper, we present a system for the movement detection of crowd gathered in video sequences. System incorporate sestimating optic flow approach and is designed to track crowd motion pattern which is a feature for harmful congestion. It follows assumption that, sequences of ordered frames of video allows the computation of motion as either discrete image or frame displacements or instantaneous image or frame velocities. we present the location of overcrowding which helps in taking further protective measures to handle unusual hazardous incidents.

LITERATURE REVIEW
Nowadays, research on crowd dynamics topic in computer vision field is gaining huge popularity due to its significant use in various applications that detects and provides early indications of densely crowded area and also supports emergency security aspects [3]. Computer vision is the core technology which deals with automated vision analysis. Thus, from several recent decades visual systems are focusing on modeling and interpreting various sorts of pedestrian activities including the individual's behavior in crowds. Pedestrian movement analysis is an active research topic from previous two decades. Earlier, dense pedestrian group and their self-behaving characteristics in crowd were studied using simulations. For identifying people behavior in densely packed crowd different physical models exist which depicts pedestrian motion as a model. They are modeled based on the correspondence to fluid dynamics. Later various popular models came into existence termed as socio-force model [4,5] and cellular automaton [6,7], which models pedestrian motion on a microscopic level.
From past two decades, various vision analysis and crowd analysis techniques have been developed specifically for tracking and recognizing unusual human activities in crowd [8,9]. Presently modeling of efficient and accurate vision-based systems embeds different models of simulation that are capable of tracking individual movement in immense gatherings. Cellular automata model admired by Ali and Shah [3] took the initiative to present a framework for detecting human in the crowd [4]. It is an automated model composed with the knowledge on human activities in crowd, locations of obstructing objects and entry or exit door's position and mainly used for computation of flow vectors. Ali and Shah also presented a framework that deals with segmentation of the pedestrian flow that helps in tracking varying patterns among crowd scenes [5].
To track unusual events in mass gatherings Mehran et al. considers social force model and estimates force interactions [6]. Further, various research and experiments are performed in crowd dynamics to understand hazardous pedestrian behavior for enhancing the existing physical methods. Numerous attributes like crowd flow, pedestrian velocity, pressure, people density are calculated either ancient known methods [7] or by means of recent automate ddigital visual processing techniques [8]. Usually, vision analysis experiments are performed on videos captured using top and front-view cameras mainly to avoid object obstructions and for supporting automated vision analysis.
Common approaches to automated video surveillance tracks and recognizes single person or object in video. Nevertheless, many works have reported the infeasibility of those approaches and thus they are not efficient for video surveillance of densely packed crowd, because huge gatherings video captured from any source will have numerous persons moving which necessitates complex calculations. Therefore, to ascertain crowd motion patterns and their directions computation of dense optic flow vectors is considered as efficient. By utilizing optical flow estimation approach not only accuracy of complex calculations is met but also the confidentiality of human identity is guaranteed. An observable motion of any object, areas or regions, edges in a video sequences that occurs due to the instantaneous movement between perceiver and the scene is referred as anoptical flow. They can also be termed as an optic flow. American psychologist James J. Gibson during the year 1940 demonstrated the concept of optical flow in order to present the idea of visual stimulus those were experienced by variety of animals roaming in any place.
Further, the importance of optic flow is also reported by James J. Gibson for explaining the concept of affordance perception, which helps in recognizing various observable activities occurring in an environment. Recently, the concept of optical flow finds immense usage in the robotics field which incorporates various combined methodologies from video or image processing as well as from navigation control, like object's motion tracking, segmentation of video frames and surface or edge detection in videos. Movement detection or estimation, video analysis or compressions are also a proactive research topic in a crowd dynamics field. Many works concludes optical flow computation knowledge is considered to be very useful in balancing the various operations of micro air vehicles. This technique can also track and inform such congested regions occurring in real-time video sequences [12][13][14] as well as video denoising technique in real time [16]. The Literature survey shows, the technique which is used in this paper is very good compared to the existing technique.

DETECTION OF OVERCROWDED REGION
In huge gatherings, people moves with decreased velocity which results in oscillatory motion. Thus, tracking their velocity and motion direction becomes much difficult because its people tendency to change directions more frequently while moving in such gatherings. Liu et al. work shows that people moves in crowd with oscillatory motion and with differing velocities from 0.26 m/s up to 1.72 m/s [14]. For study of vision dynamics various authors have reported the existence of linearity between the velocity and the amplitude also between the velocity and the frequency. We estimate dense optical flow vectors using the Lucas-Kanade approach which is a differential technique.
The perceptible movement of brightness pattern in a video sequences is called as an optical flow. Video sequences consist of ordered collection of frames or images that allow the estimation of flow vectors as either instantaneous frame velocities or discrete displacement of frames or images. Various methods exist for the calculation of optical flow vectors that detects motion between two subsequent or different frame sequences that are captured at times T and T+∆Tat every volumetric pixel location. Optic flow technique used is called differential because they consider only partial derivatives of spatiotemporal [14] co-ordinates and also approximations of frame or image sequences are done based on Taylor series assumptions.

Optical Flow Estimation Approach
The embodied Lucas-Kanade approach is a popular and robust differential method for optical flow vectors estimation. A major assumption is that the flow is essentially constant in a local neighborhood of the volumetric pixel under consideration and solves the basic optical flow equations for all the pixels in that neighborhood by making use of the least squares criterion. The Lucas-Kanade approach assumes that the displacement of the image contents between two nearby instants of subsequent frames is small and approximately constant within a neighborhood of the point punder consideration. Therefore the optical flow equation can be assumed to hold for all the pixels within a window centered at p. Thus, the local image or frame velocity vectors [m, n] must satisfy .….
where r1, r2, and rn are the pixels inside the window and are the partial derivatives of the image Iwith respect to position x, y and time t, evaluated at the point and at the current time.
To solve the optical flow constraint equation for m and n, the Lucas-Kanade approach divides the original image or frame into smaller sections and assumes a constant velocity in every section. Then, it performs a weighted least-square fit of the optical flow constraint equation and minimizes following equation: Here, W is a window function that emphasizes the constraints at the center of each section and represents corresponding section.
The stated approach solves velocity vectors m and n as follows: 1. Compute Ix and Iy using [-1 8 0 -81]/12 as convolution kernel, for every pixels in the first and second image. The optical flow, m and n, is 0.
In video analysis, a convolution kernel, or mask is a matrix used for sharpening, smoothing, edge-identification, etc. This is accomplished by means of convolution between an image and a kernel.

Segmentation of Video Sequences
For a gathered crowd video segmentation of video sequences are done for easy and efficient processing of video sequences.We decompose sequence of frames from collected crowd video and each frame or image is segmented into A p r i l 2 0 , 2 0 1 4 several parts or segments. For each segment the above mentioned optical flow computation approach is applied. In an overcrowded area, the people either move very slowly or stops moving itself. If the scenario is analyzed for top-view portion of video the estimation of crowd velocity can be done by assuming only the magnitude of optical flow vectors [15]. So for every segment flow vectors are deduced by just depending on its magnitude values. Further, a threshold optical flow vectors are chosen for comparing each segment's already estimated optical flow fields with them.
Therefore, if the optical velocity values in any segment are less than the chosen threshold then there is no crowd movement in that region and thus congestion exist. Else, if the optical flow values in any segment are more than the threshold then we conclude people in that area moves freely and thus no congestion can exist. Finally, the area is reported where there exists congestion for taking further preventive measures. This technique of estimating optical flow vectors can also be applied to real time crowd video footage for detecting congestion.

EXPERIMENTAL RESULT
The technique of movement detection is performed for human movement tracking and congestion detection in immense gatherings. We take a sample video footage of gatherings near railways and extract frames from collected video. Individual frames are segmented and for each segment optical flow is computed. Final system should be able track congested region of crowd as shown in Figure 1 where rectangular box depicts congestion area.

CONCLUSION & FUTURE WORK
The proposed system automatically detects an overcrowded area using above said optical flow estimation technique and segmenting frame sequences. It can also track and inform such congested regions occurring in real-time video sequences. The video footage assumed can be crowd at railways or overcrowding at political rally etc. Such useful information can be used to take necessary steps in controlling hazardous incidents occurring in mass gatherings.
The System can also be used to track human faces, track the noisy area of a noisy video and analyze suspicious activities in mass gatherings and inform the concerned authorities to take necessary actions.