Local invariant feature extraction, as one of the main problems in the field of computer vision, has been widely applied to image matching, splicing and target recognition etc. Lowe’s scale invariant feature transform (known as SIFT) algorithm has attracted much attention due to its invariance to scale, rotation and illumination. However, SIFT is not robust to affine deformations, because it is based on the DoG detector which extracts keypoints in a circle region. Besides, the feature descriptor is represented by a 128-dimensional vector, which means that the algorithm complexity is extremely large especially when there is a great quantity of keypoints in the image. In this paper, a new feature descriptor, which is robust to affine deformations, is proposed. Considering that circles turn to be ellipses after affine deformations, some improvements have been made. Firstly, the Gaussian image pyramids are constructed by convoluting the source image and the elliptical Gaussian kernel with two volatile parameters, orientation and eccentricity. In addition, the two parameters are discretely selected in order to imitate the possibilities of the affine deformation, which can make sure that anisotropic regions are transformed into isotropic ones. Next, all extreme points can be extracted as the candidates for the affine-invariant keypoints in the image pyramids. After accurate keypoints localization is performed, the secondary moment of the keypoints’ neighborhood is calculated to identify the elliptical region which is affineinvariant, the same as SIFT, the main orientation of the keypoints can be determined and the feature descriptor is generated based on the histogram constructed in this region. At last, the PCA method for the 128-dimensional descriptor’s reduction is used to improve the computer calculating efficiency. The experiments show that this new algorithm inherits all SIFT’s original advantages, and has a good resistance to affine deformations; what’s more, it is more effective in calculation and storage requirement.
Mean shift, which is widely used in many target tracking systems, is a very effective algorithm to track the target. But the traditional mean shift tracking algorithm is limited to track an infrared small target. In infrared prewarning and tracking systems, the traditional mean shift tracking algorithm cannot achieve accurate tracking result due to that the target is weakened and submerged in the background noise. So in this paper, a compositive mean shift algorithm is put forward. In this algorithm, firstly on the basis of background suppression and division, noise is suppressed by an extraordinary Robinson Guard Filter. This paper adopts a dual patterns merging Robinson Guard Filter which is different from the traditional Robinson Guard Filter. According to the point target’s anisotropic singularity in space, this dual patterns merging Robinson Guard Filter can divide the direction further and detect singularity accurately in different directions in order to obtain better effect. The dual patterns merging Robinson Guard Filter’s improvement is that it adopts the horizontal and vertical direction window and the diagonal direction window whose protective belt width are both two at the same time to increase the probability of point target detection. The filter separately detects the two directions and merges the results in order to boost the effect of keeping back the details of the target. At the same time, it can also boost the effect of background suppression as much as possible and reduce the false alarm rate. At last the system can achieve ideal detection performance. After filtering, an image in which the point target and the background are distinguished is acquired. Then in the mean shift algorithm, we use the acquired image for target tracking. The results of experiment show that this improved mean shift algorithm can reduce failure probability of prewarning and track infrared small targets steadily and accurately.
With the rapid development of computer technology and mobile devices, high requirements for computer vision are more and more critical. In particular, current research on moving target detection based on still scene or hardware platform with turntable and gyroscope, can not meet the requirements of portable mobile equipment. Moving target detection and tracking on mobile platforms is attracting more and more attention. What makes the task even more challenging is when the camera is non-stationary ,due to the random motion of camera caused by bumps and swings of vehicle and handheld, and parallax problem caused by 3D scene. The essential problem in this case lies in distinguishing between global motion induced by camera and independent motion caused by moving targets. To solve above problems, the three-dimensional reconstruction method based on camera calibration technology is always introduced, such as the fundamental matrix and trifocal tensor, but that are appropriate for large classes of problems and situations without considering the complexity and speed of processing. This paper proposes a new robust algorithm GMOS (Global Motion Of Scene) based on the global motion of scene for moving object detection on a freely moving camera. By modeling for GMOS with the adjacent optical flow field of the image sequence, the proposed method is able to detect and separate the moving targets simply and fast from the global motion model without three-dimensional reconstruction. According to the GMOS model, we can describe the movement of the camera through a GMOS vector which is independent of two viewpoints between the adjacent images, and compensate for the overall movement caused by camera movement. The results of theory analysis and experimentations on numerous real world videos demonstrate that the proposed method GMOS could separate the independent objects fast and robustly under the premise of the high accuracy and robustness.
It introduces a new method to achieve the
passenger flow statistics in stereo vision according
to the original depth image output by the monocular
Xtion sensor, aiming at the problem of algorithm with
large amounts of data and realization of single field
with dual camera on the basis of stereo vision.
Double Xtion sensors are used to expand the range
of view angle because of the monocular Xtion
sensor’s limitations, whose view range is 45°*58°
with small transverse view range and can’t meet the
passenger flow statistics. Due to the characteristics
of constant physical space dimensions, use the
improved SIFT (Scale Invariant Features Transform)
feature algorithm to realize the auto - stereoscopic
splice of binocular original depth images. Firstly, the
feature points of the reference image (the image to
be matched) and the subsequent image (the image
to be matched with the reference image) are
obtained by SIFT algorithm, getting the location,
scale and direction of the feature points and the
feature points are described by means of the
128-dimensional vector .Secondly, complete the
match of the feature points of the two images to
calculate overlapping area, using the nearest
neighbor method. Finally, image stitching is
completed based on multi-resolution wavelet
transform, which contains three-dimensional spatial
information of the human body, thus use a method to
analysis comprehensively the depth image for field
detection and tracking based on the features such as
the head shape, the head area the spatial position
relation of the human head and shoulder and so on.
The experimental results show that this method not
only improve the detection accuracy and efficiency,
reduce the amount of operation data, so that the
system is simple in structure, but also solve many
problems of passenger flow statistics based on video
stream in the system, accuracy up to 93%, having
high and practical application value.