A survey of video target tracking algorithms

Video Tracking : Target Tracking based on contrast analysis, target tracking based on matching and target tracking based on motion detection
target tracking based on contrast analysis : The target detection and tracking is realized by the contrast difference between target and background. This kind of algorithm can be divided into edge tracking # centroid tracking and center of mass according to the different tracking reference points.

Tracking and so on. This kind of algorithm is not suitable for target tracking in complex background, but it is very effective in target tracking in aerial background.

target tracking based on matching : The target positioning is achieved mainly through the feature matching between the front and back frames.

feature matching: The characteristic is the attribute of the target distinguishing and other things, which is distinguishable, reliable, independent and sparse. The target tracking algorithm based on matching needs to extract the characteristics of the target and look for it in each frame.

The process of finding is the process of feature matching. The features used in target tracking include geometric shape, subspace feature, contour and feature point, and feature points are common features in matching algorithm. The change of the target feature

The stochastic nature of the. This stochastic variation can be described by means of statistical mathematics. Histogram is the natural statistic in image processing. Therefore, the color and edge direction histogram is widely used in the tracking algorithm.

Bayesian Tracking: The motion of a target is often random, and such a motion process can be described by a stochastic process. The processing of stochastic process is more mature in the field of signal analysis, and its theory and technology (such as Bayesian filtering) can be used for reference to target tracking.

Nuclear method: The basic idea of the nuclear method is to use a direct continuous estimation of the similarity probability density function or the posterior probability density function, so that the processing can simplify the sampling, on the other hand, the estimated function gradient can be used to effectively locate the sampled particles.

Using the continuous probability density function can reduce the computational problem caused by the high dimension state space, also can ensure the example approaches the distribution pattern, avoids the particle degradation problem, the kernel method generally uses the color histogram as the matching characteristic.

target tracking based on motion detection : Based on the difference between target movement and background motion, the target detection and tracking are realized. The target tracking algorithm based on motion detection detects the regions where the target exists by detecting the different motions of the target and the background in the sequence image.

Implementation tracking. This kind of algorithm does not need the pattern match between frames, does not need to pass the target motion parameter between the frame, only needs to highlight the target and the non-target in the time domain or the airspace difference. These algorithms have the ability to detect multiple targets and can be used for multi-target detection

Detection and tracking, the method of detecting motion targets includes inter-frame image difference method, background estimation method, energy accumulation method, field estimation method and so on.

Optical flow algorithm is a representative algorithm of target tracking based on motion detection, and the optical flow is the instantaneous velocity of the pixel motion of the space moving object on the imaging plane, the light vectoring amount is the instantaneous rate of change of gray level on the plane coordinate point of the image, and the optical flow is computed using the image sequence.

The temporal changes and correlations of the pixel grayscale distributions in the pixels to determine the motion of their respective pixel positions. This paper studies the relationship between the change of image grayscale in time and the structure and motion of objects in the scene, and the relation between the two-dimensional velocity field and gray scale. The optical flow constraint equation is introduced,

The basic algorithm of optical flow calculation is obtained. According to the different calculation methods, the optical flow algorithm can be divided into gradient-based method, based on matching method, energy-based method, phase-based method and neuro-dynamic method.

The first two methods are the single frame image processing, based on the matching tracking method needs to transfer the target information between the frame and the frame, the contrast tracking does not need to pass the target information between the frame and the frame, and the motion detection based tracking needs to process the multi-frame image.

Another reference: http://www.cnblogs.com/zjb0823/p/3806333.html

A survey of video target tracking algorithms