Anda di halaman 1dari 5

IJIRST International Journal for Innovative Research in Science & Technology| Volume 3 | Issue 09 | February 2017

ISSN (online): 2349-6010

Advanced Real Time Objects Detection and

Tracking in HEVC Videos
Siddappa Pulare Dr. T S Indumathi
PG Student Professor
Department of Digital Electronics & Communication Systems Department of Digital Electronics & Communication Systems
VIAT Muddenhalli, Chikkaballapur-562101 India VIAT Muddenhalli, Chikkaballapur-562101 India

In this paper, we propose a modern method of tracking in which it detects multiple moving objects in videos compressed per
H.265/HEVC standard. Our proposed paper work largely depends on motion vectors (MV) and block types obtained by partially
decoding the video bit stream and occasionally uses pixel domain information to distinguish between two objects. In this work an
approach to detect the moving object in compressed domain HEVC is discussed which is used in video surveillance system.
Because High Efficiency Video Coding (HEVC) has a large potential to identify events by reusing coding structures in HEVC,
which saves vast extent of computational resources. Here the feature Motion Intensity Count (MIC) makes use of motion vector,
coding unit and prediction unit of HEVC. Our extensive experimental results demonstrate that the approach provides the
remarkable performance and can classify moving person and vehicles accurately and robustly.
Keywords: Object Tracking, Object Detection, HEVC, Classification, MIC


Multiple object detection is one of the vital technologies for intelligent vision applications such as Advanced Object Detection
and tracking Systems, mobile robot vision, surveillance systems and portable devices [1] [4]. These applications and others
require real time processing and comparatively low power consumption. Real time processing with high frame rates is essential
in proposed pater and UAVs, where the detection must be fast enough to allow sufficient time for any course correction. For
robustness, multiscale detection is important to detect objects with different sizes and distances from the objects
height are inversely proportional to its distance from the camera [5]). Processing high resolution images, such as full high
definition (HD) 1920 1080, is required to have enough pixels for early detection of far and small objects.


Now this chapter explains some of the motivations underlying the multiple object detection based video coding approach. Maanu
Tam, R. Venktesh Babu [5], they both proposed a method for moving object detection in H.264/. AVC standard compressed
domain. The proposed algorithm is mainly initially Segments based up on the quantization parameter and macro block size, their
proposed approach allows the video stream to be encoded with various different quantization parameters across macro blocks
there by increasing flexibility in bit rate adjustment. C. Popee [5], they proposed a fast method for motion segmentation for
H.264/. AVC based up on the size and transform coefficients used within the video stream. This algorithm needs the values of
QP to be fixed while encoding, reducing bit rate flexibility.
Kalyani Goswami [6], she proposed a new mesh based moving object detection technique for tracking. A dense mesh has been
generated on that part of a frame where the moving object are present else less dense mesh has been formed. This algorithm does
not require any prior information about the location of the object inside a video frame. This algorithm captures moving object
inside a video sequence without any pre- or post-processing. The motion characteristics are tuned manually. Dingaming Liua
[4], Here they introduced a new approach to extract the object based on moving detection and graph cut. Firstly, they get the
moving information of the object in each frame with frame difference of three continuous frames.
Here the possibility foreground frame which will segment to extract, it can be solved by graph values, extracting the object in
each frame with the graph values method to segment every frame to extract the multiple moving objects. This system still has
some limitations. Firstly, when the foreground frame and background frame colors are very similar, high quality segmentation
usually is hard to be obtained. Vijaya Jumb [7], this approach for color image frame segmentation is presented. In this algorithm
foreground frame objects are distinguished clearly from the background frames. Initially RGB is converted into HSV and extract
V from HSV, then apply Otsu on V channel and then apply K-means algorithm for over segmented.


The proposed system is having first to analyze the captured image frames to find multiple moving objects, a decoding step is
needed. By using this decoding step and we reused the work done during the encoding to detect multiple moving objects directly

All rights reserved by 40

Advanced Real Time Objects Detection and Tracking in HEVC Videos
(IJIRST/ Volume 3 / Issue 09/ 008)

upon the compressed video stream. The vital idea in our proposed system is we are presenting segmentation to extract the
multiple moving objects in HEVC standard compressed domain the general flow chart of the system is illustrated in following
proposed diagram (2.1).

Fig. 1: Block diagram of background subtraction method

In this block diagram, there are two prominent things first one is reference image, it is for the reference while detecting object
this play a prominent role and second one is input image it is for except reference image in the video frames. the difference
between input image and reference image which gives a difference image so that with help of this we can detect an object easily,
so this method is called as a background subtraction method .by thresh holding and compensating we can detect exact objects
with more accuracy for that dilation, err ration, opening and closing morphological image processing will gives an accurate result
in detecting objects by background subtraction method. The above explanation block diagram is as shown below figure (2.2).

Fig. 2: The system block diagram

As we firstly going through the related standard compressed domain information in H.265 standard. In HEVC, all frame is
firstly divided into squares of same size, i.e. largest coding units. Every LCU can be recursively divided into smaller unit until a
proper partition reached. Every smallest unit is referred to as prediction unit (PU). For each inter-coded PU, there will be a
motion vector (MV) indicating the best match position in the reference frame. Thus, to represent the motion intensity within a
LCU. MIC for each LCU is defined as follows: Using ||MVi|| the set of the MVs within the same LCU, area(PUi)is the is the area
of the PU containing this MV. LCU with larger MVs will have higher MIC.
Based up on the features like MIC and the MIC prediction method, as we multiple moving objects by identifying the
differences between the predicted and actual MICs for every frame. For every currently checking frames, we predict MICs for
current frame by previous frames, which are inside a sliding window of size s. For each LCU in current frame, with the predicted

All rights reserved by 41

Advanced Real Time Objects Detection and Tracking in HEVC Videos
(IJIRST/ Volume 3 / Issue 09/ 008)

MICs and actual MIC we can measure the differences. If difference is greater than zero, then there is possibility of moving
objects are present in the LCU. Apply adaptive thresholding method on LCU whose value of d is greater than zero to extract the
moving objects.
The global threshold value is not good in all the conditions where image has different lighting conditions in different areas. In
that case, we go for adaptive thresholding.
Here we calculate the threshold for a small region of the image. So, we get different thresholds for different regions of the
same image and it gives us better results for images with varying illumination


The proposed system can be viewed as an important step in moving forward towards having outperforming methods in the field
of real-time multiple object detection and tracking. The main involvement in this system is the presence of a new architecture for
processing images using background subtraction algorithm along with HEVC video compression technology gives robust to this
project and which made this project more robust system which benefits from the combination between HEVC/H.265 video
compression standard method and the detection and tracking techniques such as the background subtraction method and some
morphological image processing. Using the parameters found in the system design, we tested our MATLAB software using real
time HEVC videos with different instances for each image. Our software/CODE can detect objects with different orientation,
scaling and transformation.
The below fig (4.1) contains three windows are listed below
Original Video
Binarized Difference video
Tracking vehicle

Fig. 3: Basic window developed by MAT Lab code

All rights reserved by 42

Advanced Real Time Objects Detection and Tracking in HEVC Videos
(IJIRST/ Volume 3 / Issue 09/ 008)

Fig. 4: Vehicle detected and tracking

Fig. 5: The proposed system detecting tracking and classification and counting in same window


Multiple Moving objects identification is based upon the information of HEVC standard compressed domain feature in which is
Motion Intensity Count (MIC) which is extracted from compressed domain information of HEVC frames and the Motion
vectors, coding unit and prediction unit are used to identify MIC. Then the vital role of concrete segmentation is carried by
Adaptive thresholding to identify the moving objects. Here adaptive thresholding is applied only on the largest coding unit
whose value of difference in MIC values is greater than zero, but not on entire frame, this is important where image has different
lighting conditions in different areas and it will improve performance simply by applying Adaptive Otsu on specific LCU. This
system will be useful for video surveillance applications for example like detecting severe traffic incidents online and make it

All rights reserved by 43

Advanced Real Time Objects Detection and Tracking in HEVC Videos
(IJIRST/ Volume 3 / Issue 09/ 008)

possible to warn traffic operators to take urgent action etc. Compared to other systems, the proposed approach allows the video
streams to identify moving object in different approach i.e., by using MIC.

[1] Huang LI, Yihao ZHANG, Ming YANG, Yangyang MEN, Hongyang CHAO, A rapid abnormal event detection method for surveillance video based on
a novel feature in compressed domain of HEVC, 2014 IEEE International Conference on Multimedia and Expo (ICME), 14-18 July 2014.
[2] H.T. Chen, H.H. Lin, and T.L. Liu. Multi-object tracking using dynamical graph matching. In Proc. of IEEE Computer Society Conference on Computer
Vision and Pattern Recognition, pages 210217, 2001.
[3] Vinutha H Reddy, Prof. Chhaya S. Pawar, Paper Published in International Journal of Research in Advent Technology, Volume 3, Issue 3, March 2015
Moving Object Detection in Compressed Domain of HEVC for Video Surveillance
[4] C. Popee, S. D. Bruyne, T. Paridaens, P. Lambert, and R. Walle,Moving object detection in the h.264/avc compressed domain for video surveillance
applications, Journal of Visual Communication and Image Representation, vol. 20, no. 6, pp. 428437, 2009.
[5] Kalyani Goswami, Gwang-Soo Hong, Byung-Gyu Kim A Novel Mesh-Based Moving Object Detection Technique in Video Sequence, Journal of
Convergence Volume 4, Number 3, September 2013.
[6] R. V. Babu, K. R. Ramakrishnan, and S. H. Srinivasan, Video object segmentation: a compressed domain approach, IEEE Transactions on Circuits and
Systems for Video Technology, vol. 14, no. 4, pp. 462474, April 2004.
[7] R. Bodor, B. Jackson, and N. Papanikolopoulos. Vision-based human tracking and activity recognition. In Proc. of the 11th Mediterranean Conf. on Control
and Automation, June 2003.
[8] M. Pedersoli, J. Gonzlez, X. Hu, and X. Roca, Toward Real-Time Pedestrian Detection Based on a Deformable Template Model,IEEE Transactions on
Intelligent Transportation Systems, vol. 15, no. 1,pp. 355364, Feb. 2014
[9] H. Sharma, T. Dutta, V. Adithya, and P. Balamuralidhar, A Real-Time Framework for Detection of Long Linear Infrastructural Objects in Aerial
Imagery,International Conference on Image Analysis and Recognition(ICIAR), pp. 7181, July 2015.
[10] Philipp Wojaczek, Marcus Laumer, Peter Amon, Andreas Hutter, and Andre Kaup, Hybrid Person Detection and Tracking in H.264/AVC Video
Streams,Proceedings of the 10th International Conference on Computer Vision Theory and Applications, pp. 478485, 2015

All rights reserved by 44