Summarization of algorithms for moving target detection and tracking process

Summarization of algorithms for moving target detection and tracking process

Image preprocessing Some typical noises in digital images are: Gaussian noise comes from the noise of electronic circuit and low illumination or high temperature. The salt and pepper noise is similar to that of pepper and salt powder randomly distributed on the image, which is caused by the image cutting and the error caused by the transform domain; additive noise is the channel noise that the image introduces in the transmission. Generally speaking, the introduced is additive random noise, can use mean filter, median filter, Gaussian filter and other methods to remove the noise, improve the signal-to-noise ratio. The mean filter can be used better when the noise distribution is more average and the peak value is not very high; The median filter has a good effect on the filtering of the sharp pulse noise, and it can highlight the edges and details of the image. Gaussian filtering has a good effect on filtering Gaussian white noise. Moving target detection Background Difference method: The motion object can be segmented quickly and completely. Deficiencies are susceptible to changes in light, and background updates are key. does not apply to camera motion situations. Optical Flow method: can detect independent motion of the object, can be used for camera motion situation, but the complexity of the calculation time-consuming, difficult to detect in real time. Frame Difference Method: The effect of light changes is small, simple and fast, but can not be divided into a complete moving object, the need to further use of target segmentation algorithm. There are also improved algorithms that focus on reducing lighting effects and detecting slow object changes. Image Identification The role of image identification is to determine whether the object is independent, there are several moving objects in the image. 1) field: Always take the surrounding 4 or 8 pixels as a field. 2) Connected domains: Two-value images with interconnected 0-pixel sets or 1-pixel sets called connected domains. 0 pixels surrounded by 1 pixels are called holes. When a 1-megapixel connected domain is free of holes, it is called a single-connected component, with one or more holes connected as a multi-connected component. 3) Mark: After the difference of a frame image may exist multiple connected domains, each non-connected domain corresponding to a target image area, the target area assigned to the corresponding label work becomes a marker. The identification process is roughly: scan pixels in a certain order, scan to 1 of pixels, detect pixel values in their domain, and, if the same, connect to the domain, mark as the first target, and then look for the next one in turn. After all possible targets have been found, you can draw a gate for each target and frame the target. and set up a multi-target location linked list, find the central location of each target area to join the list as a node to store it. The division of the gate may be divided into two parts of the same target, or a gate contains two targets, so that the target data errors increase or decrease, so also to determine whether the current target is the same target or different targets, which will be completed in the image segmentation later. Image Segmentation Image segmentation is used to separate the combination of target and background or to separate different objects. Image segmentation not only can compress data, reduce storage capacity, but also greatly simplify the subsequent analysis and processing steps. 1) Histogram threshold segmentation method Gray-scale histogram is a two-dimensional relationship between the number of pixels and gray scale, which reflects the statistical characteristics of an image gray distribution. If the gray value distribution of the foreground object is more uniform, the distribution of the background gray value is more uniform, the histogram of this image will have obvious Shuangfeng, then the valley between two peaks can be chosen as the threshold value. Because the histogram does not contain the location information of the target, it should be determined by combining the contents of the image. 2) Maximum inter-class variance threshold Segmentation method Based on the difference between the image target and the background, the gap between the two groups is used to determine the threshold value for segmentation. 3) Regional Growth method Refers to the merging of pixels with similar surrounding characteristics into the target area again. 4) edge detection and Contour extraction segmentation method 5) Morphological Segmentation method The main function is to make the area of the moving target more complete. The effect of corrosion is to eliminate the object boundary point and remove objects less than the structural elements. If there is a small connection between the two objects, when the structure element is large enough, two objects can be separated by the corrosion operation. The function of the expansion operation is to merge the background points around the image into the object. If two objects are closer, the expansion operation may cause the two objects to connect together. Expansion is useful for filling holes in images, and one of the simplest applications of swelling is to connect the cracks together. Morphology can also be used in image filtering, enhancement and so on. Motion trajectory Prediction After dividing the moving target, the characteristics of the target should be extracted, and then match the feature in the next frame image to track the target. However, in order to reduce the area of search feature matching and improve the real-time, this step is to predict the target motion trajectory. Motion trajectory prediction is also helpful to enhance the robustness of tracking in occlusion case. 1) Linear prediction algorithm 2) Kalman filtering algorithm and its extension algorithm 3) Particle filter algorithm Target Tracking 1) Feature Selection Gray-scale feature, for gray-scale image, the pixel gray value is the most basic target image feature; Geometric characteristics, which reflect the geometric nature of the target, which is related only to the location of the target pixel, and not to its grayscale. The common geometric features include target circumference, area, flattening rate and height, etc. Statistical characteristics, such as the target gray mean and variance, histogram, entropy, moment and the target relative to the background contrast; Transform domain features, including forier, Gabor, wavelet and other transformation domain features; Color characteristics. 2) Tracking algorithm A. Template matching tracking The template slides on the image, corresponds to the gray value of each position of the image, compares with the gray value of the pixel on the template, calculates a cumulative error for each position, selects the most suitable position, and completes the match. B.camshift Tracking Color histogram matching. C. Active contour Line Tracking Also called Snake algorithm, minimizing the energy function of the image. D. Multi-mode tracking Track simultaneously using multiple tracking algorithms (models). the complexity of moving target detection and tracking is manifested in 1, light changes. Due to time changes, day and night, morning and afternoon, daylight intensity and angle changes will cause light changes, and due to weather conditions, will also lead to light changes. Due to the change of light, it is difficult to adapt a detection algorithm to all kinds of illumination conditions. 2. The disturbance of moving object in the scene. For example, the movement of various objects in a large area, sudden stop and sudden start of vehicles, frequent changes in the scene of certain objects, such as swaying branches and leaves, fluctuations in the surface of the water, etc. 3, the initialization problem. In some monitoring scenarios, it is difficult to obtain a pure background image without noise interference (images without detection targets and moving background targets). For example, people's cars are busy traffic scenes. 4, occlusion and hole problems. To detect the moving target, is obscured by the target in the background, how to judge the occlusion. 5. Shadow problem. In target detection, how to distinguish the detected target from the shadow it produces, so that only the target part is detected. 6, the disappearance of the target. Moving objects remain in the scene for a long time and are likely to become background targets. Because of the complexity of target detection, it is unrealistic and impossible to build a generic, target detection algorithm for all situations. Therefore, according to the specific situation, it is the research direction of target detection method to establish the target detection algorithm which accords with the actual condition.

Http://hi.baidu.com/gilbertjuly/blog/item/be956217178f8712962b43ce.html

Summarization of algorithms for moving target detection and tracking process