 Our Main aim is to detect the Object in the running

videos for the purpose of security enhancement.

 The performance of a generic Object detector may drop
significantly when it is applied to a specific scene due to the
mismatch between the source training set and samples from
the target scene.
 We propose a new approach of automatically transferring a
generic Object detector to a scene-specific detector in static
video surveillance without manually labeling samples from the
target scene.
 The proposed transfer learning framework consists of four
steps.
 Object detection is of great interest in video surveillance.
 Many existing works are based on background subtraction, which is
sensitive to lighting variations and scene clutters, and has difficulty
in handling the grouping and fragmentation problems.
 Most surveillance cameras are stationary. If a scene is fixed, the
variations of positive and negative samples will be significantly
reduced, since videos captured with a single camera only have
limited variations on viewpoints, resolutions, lighting conditions,
and backgrounds.
 Sensor based Human Identifications are there.
 Existing system uses fixed surveillance camera but that
will failed to differentiate between positive and
negative sample.
 It having limited variations on viewpoints, resolutions,
lighting conditions, an backgrounds.
 Accuracy was low when compare to proposed system.
 And it was not discussing about contrast problems.
 Unnecessary Power Losses, And expensive.
 We have introduced a weighted soft margin SVM to
incorporate prior knowledge in the training process.

 we have proposed a transfer learning approach to adapt

weights of weak classifiers learned from a source data set to
a target data set to handle the variation of viewpoints.

 It assumes that some samples in the target set are manually

labeled and does not consider context cues.
 When cameras are stationary, the distributions of negative
samples are region-specific.

 We have trained a separate detector for each local region.

have used tracking and manually input scene geometry to
assist labeling.

 have proposed Flow Boost to learn a scene-specific detector

from a sparsely labeled training video assisted by tracking.
 1) Through exploring the in degrees from target samples to source samples on a
visual affinity graph, the source samples are weighted to match the distribution
of target samples.
 2) It explores a set of context cues to automatically select samples from the
target scene, predicts their labels, and computes confidence scores to guide
transfer learning.
 3) The confidence scores propagate among target samples according to their
underlying visual structures.
 4) Target samples with higher confidence scores have larger influence on
training scene-specific detectors.
 CONVERTING VIDEOS INTO FRAMES
 BACKGROUND SUPRESSION
 INCRESING THE PIXEL DENSITY
 FEATURE EXTRACTION
 DATABSE STORING
 COMPARE THE DATA
 DECISION MAKING
 Contrast problems was overcome
 PSNR value was low in proposed system
 Running Videos also effectively achieved.
 High frame rate images also processed effectively
 System : Pentium IV 2.4 GHz.
 Hard Disk : 80 GB.
 Monitor : 15 VGA Color.
 Mouse : Logitech.
 Ram : 512 MB.
 Operating system : Windows (64-Bit)
 Front End : MATLAB
 Coding Language : Java, MATLAB C
 In this paper, four typical LC accident scenarios acquired in real conditions have

been experimentally evaluated by applying the proposed dangerous situation

recognition system.

 A risk index has been defined to assess the risk of objects detected in LC

environment.

 The method starts by detecting and tracking objects seen in the monitored zone by a

video camera.

 The second stage of the method consists in predicting for each tracked object the

ideal trajectory allowing to avoid potential dangerous situations.

 The ideal trajectory prediction is based on an SVM.

 [1] A. Nelson, “The UK approach to managing risk at passive level crossings,” in Proc.
7th Int. Symp. RailRoad-Highway Grade Crossing Res. Safety, 2002, pp. 1–11.

 [2] E. Griffioen, “Improving level crossings using findings from human behaviour
studies,” in Proc. 8th Int. Level Crossing Symp., Sheffield, U.K., 2004.

 [3] [Online]. Available: http://www.uic.org/cdrom/2008/11_wcrr2008/pdf/G. 2.4.5.4.pdf

[4] L. Khoudour, M. Ghazel, F. Boukour, M. Heddebaut, and M. El-Koursi, “Towards
safer level crossings: Existing recommendations, new applicable technologies and a
proposed simulation model,” Eur. Transp. Res. Rev., vol. 1, no. 1, pp. 35–45, Mar. 2009.

 [5] J. J. Garcia et al., “Efficient multisensory barrier for obstacle detection on railways,”
IEEE Trans. Intell. Transp. Syst., vol. 11, no. 3, pp. 702–713, Sep. 2010.