Motion analysis based moving object detection from UAV aerial image is still an unsolved issue due to inconsideration of proper motion estimation. Existing moving object detection approaches from UAV aerial images did not deal with motion based pixel intensity measurement to detect moving object robustly. Besides current research on moving object detection from UAV aerial images mostly depends on either frame difference or segmentation approach separately. There are two main purposes for this research: firstly to develop a new motion model called DMM (dynamic motion model) and secondly to apply the proposed segmentation approach SUED (segmentation using edge based dilation) using frame difference embedded together with DMM model. The proposed DMM model provides effective search windows based on the highest pixel intensity to segment only specific area for moving object rather than searching the whole area of the frame using SUED. At each stage of the proposed scheme, experimental fusion of the DMM and SUED produces extracted moving objects faithfully. Experimental result reveals that the proposed DMM and SUED have successfully demonstrated the validity of the proposed methodology.
Moving object extraction observed in a video sequence and estimation of corresponding motion trajectories for each frame are one of the typical problems of interest in computer vision. However, in real environments moving object extraction becomes challenging due to unconstraint factors that is, rural or clutter environment, brightness or illumination, static or dynamic object types which together motion degeneracy may result in worthless for moving object extraction [
Motion carries a lot of information about moving object pixels which plays important role for moving object detection as image descriptor to provide a rich description of object in different environment [
This paper presents moments based motion parameter estimation called dynamic motion model (DMM) to limit the scope of segmentation called SUED which influences overall detection performance. Based on analysis from previous moving object detection frameworks, this paper used DMM model which is embedded under frame difference based segmentation approach and can handle robustness for optimum detection performance.
For accurate detection, motion must be accurately detected using suitable methods which are affected by a number of practical problems such as motion change over time and unfixed direction of moving object. Motion pattern analysis before detecting each moving object has started to get attention in recent years, especially for crowded scenarios when detecting each individual is very difficult [
As the scene may contain different motion patterns at one location within a period of time, that is, road intersection, averaging or filtering before knowing the local structure of motion patterns may destroy such structure. This paper proposes to use effective motion analysis for moving object extraction. In Figure
Existing approaches for motion based object detection.
Global illumination compensation approach works on brightness or illumination changes. Due to dependency on brightness, in real world this research does not progress so far. In parallax filtering approach, a scene that contains strong parallax is still difficult for existing methods to achieve good segmentation results [
Table
Comparison of parameters for various motion analysis approaches.
Parameters | Approaches | |||
---|---|---|---|---|
Illumination compensation | Parallel filtering | Contextual information | Long term motion analysis | |
Strong parallax situation | No | No | No | Yes |
Level of motion detection | Low | Low | Low | Low and high |
Environmental condition | No | No | Yes | Yes |
Lot of parameters | Yes | Yes | Yes | No |
Computational complexity | High | High | High | High |
Indeed, detection of motion and detection of object are coupled. If proper motion detection is done, detection of moving object from UAV aerial image becomes easier. Very few researches concentrate on adaptive robust handling of noise and unfixed motion change as well as unfixed moving object direction. For that reason an adaptive and dynamic motion analysis framework is needed for better detection of moving object from UAV aerial images where overall motion analysis reduces dependency on parameter. In other words, detection of motion indicates detection of motion pixels from frames which can be described as some function of the image pixel intensity. Pixel intensity is nothing but the pixel color value. Moments are described with respect to their power as in raised-to-the-power in mathematics. Very few previous researches used image moments to present motion analysis. Thus, this paper proposes to use image moments before segmenting individual objects and to use motion pattern in turn to facilitate the detection in each frame.
Moving object detection from UAV aerial images involves dealing with proper motion analysis. Previously very few researchers used methods which involve effective motion analysis. In [
Among these methods only frame difference approach is involved with motion analysis although most of previous research does not provide proper motion estimation to handle six uncertainty constraint factors (UCF) [
Detection rate [
This paper states that as frame difference cannot obtain motion for the complete object alone and segmentation does not have the ability to differentiate moving regions from the basic static region background, so applying frame difference and segmentation together is expected to give optimum detection result with high detection speed for moving object detection from UAV aerial images instead of applying frame difference or segmentation separately. For that reason this paper proposes moments based motion analysis to apply under frame difference based segmentation approach (SUED) which ensures robustness of the proposed methodology.
Proposed moments based motion modeling is depicted in Section
In computer vision information theory, moments are the uncertainty measurement of the object pixels. Besides, an image moment is a certain particular weighted average (moment) of the image pixel’s intensities, or a function of such moments, usually chosen to have some attractive property or interpretation. Image moments are useful to describe objects after segmentation. Properties of the image which are found via image moments are centroid, area of intensity, and object orientation as shown in Figure
Properties of image moments.
Each of these properties needs to be invariant by the following terms: translational invariant, scale invariant, and finally rotation invariant. Any feature point is called translational invariant if it does not distinguish under different points in space. Or translational invariant means that a particular translation does not change the object. Scale invariance is a feature of objects or laws that does not change if scales of length, energy, or other variables are multiplied by a common factor. The technical term for this process is known as dilation. A feature point is said to have rotation invariant if its value does not change when arbitrary rotations are applied to its argument.
Image moments can simply be described as some function of the image pixel intensity. Pixel intensity is nothing but the pixel color value. Moments are described with respect to their power as in raised-to-the-power in mathematics. This research calculates zeroth moment, first moment, second moment, and so forth from raw moments. Later this research transformed these moments into translational, scale, and rotation invariant. This research presents the following organized structure to calculate moments shown in Figure
Flow of moments calculation in the proposed research.
Before finding central moments it is necessary to find raw moments of
Using centroid coordinates, central moments for
For
Output of calculated moments proposed in this research is depicted in Figures
(a) Search window 1 for
Before going to newly proposed SUED algorithm using DMM model, input frame needs to be projected on the search (black box) only to reduce the risk of extraction failure. This research emphasizes that using a search window around the original object is very important. It limits the scope of segmentation to a smaller area. This means that the probability of the extraction is getting lost because a similar coloured object in the background is reduced. Furthermore, limited area increases the processing speed, making the extraction very fast.
This research used morphological dilation to ensure that extracted moving region contains moving objects. The output of morphological dilation is an edged image. First each frame is decomposed into
(a)
Let
(a) Difference pixel structure of
Equation (
(a)
This research proposed the following segmentation approach: segmentation using edged based dilation (SUED) algorithm for moving object extraction from UAV aerial images after extracting frame from DMM approach. Start End.
For the experiment purpose this research used IMAGE PROCESSING LAB (IPLAB) available at
Let
This research used two UAV video datasets (actions1.mpg and actions2.mpg) from Center for Research in Computer Vision (CRCV) in University of Central Florida (
This research extracted 395 frames using 1 frame/second video frame rate from actions1.mpg video datasets and 529 frames using same frame rate from actions2.mpg video datasets. Frame size is
Figures
Moments measurement from different search windows for SUED.
Search window | Zeroth moment | First moment |
Second moment |
Second moment |
---|---|---|---|---|
1 |
|
|
|
|
|
|
|
|
|
3 |
|
|
|
|
In Table
3D line chart for moment’s measurement.
This section presents some of the experimental analysis and results for the proposed SUED algorithm. The evaluation of the proposed approach was tested on actions1.mpg and actions2.mpg video analysis. In order to evaluate SUED algorithm, two metrics, detection rate (DR) and the false alarm rate (FAR), are defined. These metrics are based on the following parameters. True positive (TP): detected regions that correspond to moving object. False positive (FP): detected regions that do not correspond to a moving object. False negative (FN): moving object not detected. Detection rate or precision rate, False alarm rate or recall rate,
From dataset actions1.mpg, this research extracts 395 frames using 1 frame per second, and from actions2.mpg, this research extracted 529 frames using the same frame rate. Details of measurement for true positive (TP), false positive (FP), false negative (FN), detection rate (DR), and false alarm rate (FAR) are mentioned in Table
Details of measurement of true positive (TP), false positive (FP), false negative (FN), detection rate (DR), and false alarm rate (FAR).
Datasets | Number of frames | True positive (TP) | False positive (FP) | False negative (FN) | Detection rate (DR) | False alarm rate (FAR) |
---|---|---|---|---|---|---|
Actions1.mpg | 395 | 200 | 100 | 75 | 75% | 31% |
Actions2.mpg | 529 | 320 | 113 | 83 | 79% | 26% |
Detection rate increases if the number of input frames is increased. The detection rate of the given total frame for two video datasets is displayed in Figure
Detection rate or precision rate.
The false alarm rate for the given number of frame from two video datasets is given in Figure
False alarm rate or recall rate.
Recall and precision rate (RPC) characterizing the performance of the proposed research are given in Figure
RPC characterization.
This research measures detection and false alarm rate based on the number of frames extracted from each video dataset input. Compared with [
The primary purpose of this research is to apply moments based dynamic motion model under the proposed frame difference based segmentation approach which ensures that robust handling of motion as translation invariant, scale invariant, and rotation invariant moments value is unique. As computer vision leverages probability theory, this research used moments based motion analysis which provides search windows around the original object and limits the scope of SEUD segmentation to a smaller area. This means that the probability of extraction is getting lost because a similar colored object in the background is reduced. Since moments are the unique distribution of pixel intensity, so experimental result of the proposed DMM and SUED is very promising for robust extraction of moving object from UAV aerial images. Judging from the previous research in computer vision field, it is certain that the proposed research will facilitate UAV operator or related researchers for further research or investigation in areas where access is restricted or rescue areas, human or vehicle identification in specific areas, crowd flux statistics, anomaly detection and intelligent traffic management, and so forth.
The authors declare that there is no conflict of interests regarding the publication of this paper.
This research is supported by Ministry of Higher Education Malaysia research Grant scheme of FRGS/1/2012/SG05/UKM/02/12 and ERGS/1/2012/STG07/UKM/02/7.