|
|
Posted on 5/23/2019 6:54:43 PM
|
|
|
Intelligent video retrieval algorithm
Video retrieval relies on video algorithms to analyze video content, extract key information in videos, mark or process related processes, and form corresponding events and alarms monitoring methods, so that people can quickly retrieve through various attribute descriptions. If the camera is regarded as a person's eyes, the intelligent video surveillance system can be understood as the human brain. Intelligent video technology uses the powerful computing function of the processor to analyze massive data in the video screen at high speed and obtain the information people need.
Frame difference model
Frame difference can be said to be the simplest background model, specify an image in the video as the background, compare the current frame with the background, filter the small differences as needed, and the result is the foreground.
Background statistical model
Background statistical model is a method of counting the background over a period of time, and then calculating its statistics (such as mean, mean difference, standard deviation, mean drift value, etc.), and using the statistics as the background.
Code this background model
The basic idea of the encoded book is as follows: for the change of each pixel on the timeline, establish multiple (or one) boxes that include all recent changes; When detecting, the current pixel is used to compare with the box, and if the current pixel falls within the range of any box, it is the background.
Hybrid Gaussian model
Hybrid Gaussian background modeling is one of the more successful background modeling.
Why do you say that? Machine vision algorithms extract the basic problems faced by moving targets: image shake, noise interference, light changes, cloud drifting, shadows (including target shadows and object shadows outside the area), reflections inside the area (such as water surface, display), slow movement of moving targets, etc. So let's take a look, how does hybrid Gaussian background modeling solve these problems?
Through background modeling and foreground extraction, the target objects in the video frame are extracted, but the extracted are all non-background objects, that is, mixed, may include many people, cars, animals and other objects, and finally the image search to compare the similarity between the object and the search target, here it is necessary to separate these mixed objects through object detection and tracking.
In terms of object detection, the algorithms learned include Bayesian method, Kalman filter, and particle filter, and the relationship between them is as follows:
Bayesian method uses known information to establish the probability density function of the system and obtain the optimal solution for the estimation of the system state.
For linear Gaussian estimation problems, the expected probability density function is still Gaussian distribution, and its distribution characteristics can be described by means and variances, and the Kalman filter solves this kind of estimation problem well.
Particle filter - Sequence importance sampling particle filter is an analog-based statistical filter suitable for strong nonlinearity and no Gaussian constraints.
On the whole, the effect of particle filtering is better;
Lighting processing: The visual effect of the same object under different lighting is different, and the corresponding data is also different, so in order to improve the accuracy of analysis and recall, it is necessary to do lighting treatment on the target object; In terms of light processing, the more popular algorithm in the industry is the intrinsic image decomposition method;
Eigenimage decomposition
The most important information represented by the properties of each pixel value in the image obtained by the camera are brightness (shading) and albedo (reflectance). The brightness corresponds to the lighting information in the environment, and the albedo corresponds to the material information of the object, that is, the reflection characteristics of the object to the light, and the albedo is mainly the color information of the object. The problem of solving the intrinsic image is to start from the image, restore the brightness and albedo information in the scene corresponding to all pixels, and form the brightness eigenmap and albedo eigenmap respectively
The intrinsic image decomposition can be expressed as I(x,y) = L(x,y)R(x,y), where I(x,y) represents the input image, R(x,y) represents the albedo image, and L(x,y) represents the brightness image. Because in the logarithmic field, multiplication is converted into a much easier addition, we calculate in the logarithmic field of the image, with /(X,y) = log(I(x, y)), r0,y) = log(R(x,less)), l(x,y) = log(L(x,y)). Thus, the original multiplication relationship is transformed to: i(x,y,t) = r(x,y) + l(x,y,t).
|
Previous:Typescript TimeDateDate formats stringsNext:asp.net background task hangfire framework tutorial
|
Landlord|
Posted on 6/3/2019 9:17:36 AM
|
Come on