CN105208398B - A kind of method for obtaining the real-time Background of road - Google Patents

Abstract

The invention discloses a method for obtaining a real-time background image of a road. Grid clustering is used to quickly obtain a road background image. Firstly, the road is divided into many grids, and secondly, the statistics of the H components are performed on these grids, and then according to the H component curve The graph determines the clusters into which different video frames are clustered, and finally extracts feature frames from them, and finally extracts feature frames from them. The method has the advantages of low complexity, high efficiency and strong practicability.

Description

A method of obtaining real-time background images of roads

technical field

The invention belongs to the field of digital image processing and clustering, and is particularly suitable for extracting background images of road videos. The modeling of the road background image is an effective method for video compression and transmission processing, which involves a new grid clustering method to extract the feature frames of the background image.

Background technique

Recently, background modeling has played an increasingly important role in efficient surveillance video coding. At the same time, many practical video coding applications also put forward some specific requirements for background modeling, such as low storage cost and low computational complexity.

Existing background modeling methods can be roughly classified into 2 categories: parametric methods, such as Gaussian mixture model-1, Gaussian mixture model-2, Gaussian mixture model-3, and non-parametric methods including Bayesian models, kernel density estimation , temporal median filtering, mean shift, etc.

These methods have the disadvantages of many mathematical formulas, complex coding, and low operating efficiency. It is urgent to adopt new, simple and efficient methods for background modeling.

Through the search of existing patents and related technologies, it is found that the existing methods and systems related to the technology of obtaining road background images include:

⑴Low-complexity and high-efficiency background modeling for surveillance video coding, 2012IEEE International Conference on Visual Communication and Image Processing, San Jose, USA, pp.1-6, 11/2012

A division and weight-based (SWRA) method is proposed. First, the position of each pixel in the SWRA in the training frame is divided into several time periods, and then its corresponding average value and weight are calculated. Afterwards, a weighted average process is used to reduce the influence of foreground pixels and obtain modeling results.

⑵A Fuzzy Background Modeling Approach for Motion Detection in Dynamic Backgrounds, Multimedia and Signal Processing Volume 346 of the series Communications in Computer and Information Science pp 177-185

A method is proposed to use the fuzzy logic system recursive adaptive filter to calculate the updated weights in the Gaussian Mixture-2 model, and finally to obtain the road background map. The result proves that the fuzzy method has a great advantage compared with the traditional method.

⑶Difference of Gaussian Edge-Texture Based Background Modeling for Dynamic Traffic Conditions, Advances in Visual Computing Volume 5358 of theseseries Lecture Notes in Computer Science pp 406-417

A method for obtaining road background and detecting cars based on Gaussian edge texture is proposed. This method establishes connections between pixels and their edge and non-edge pixels, has strong learning performance, and can detect and classify road prospects well.

It can be seen that the above methods are all based on the Gaussian model for road map processing, which has the disadvantages of complex formulas, difficult coding implementation, and lack of real-time performance. It is not suitable for the current road monitoring field that requires high real-time performance.

In the existing patents, there is no clear method for obtaining the road background map for clustering, so the method we propose to obtain the road background map based on grid clustering has good research significance and application value.

Contents of the invention

In view of the shortcomings of the existing solutions stated above, the present invention aims to provide an efficient and simple method to overcome the above shortcomings of the prior art.

In order to achieve the above object, the consideration of the present invention is:

On a normal driving road, there are vehicles on each frame of the picture. But using the concept of a grid, each frame of video is divided into fine grids. We use the H component in HSV to make a color histogram of the video picture in each grid, and make a graph of the maximum value of all the H components of the grid within a period of time. When there is no car in the grid, the maximum value curve of the H component is basically stable. When a car passes by, especially when there is a big difference between the color of the car and the color of the background image, the maximum value of the H component will change dramatically. The color of the car is exactly the same as that of the road, or the probability of cars of the same color passing in series is basically non-existent.

Its specific treatment includes the following means:

A method for obtaining real-time background images of roads, using grid clustering to quickly obtain road background images, first divides the road into many grids, secondly performs H-component statistics on these grids, and then determines according to the H-component curve Clusters of different video frames, and finally extract feature frames from them, including the following processing methods:

(1) Extract the 30-second road video into frames, the video is 30 frames per second, a total of 900 frames;

(2) Perform grid interception and division for each frame, and each frame is divided into 100 grids, which are represented by two-dimensional matrices of A-J and 1-10;

(3) Calculate the H component in HSV for each grid of each frame, and count the maximum value of the H component of each grid; draw the maximum value of the H component in each grid of each frame of 900 frames into a graph ;The fluctuation of the curve indicates that the main color has shifted in this time period, and there are vehicles passing through the area;

(4) Extract feature frames by clustering method:

a. Start calculation from grid A1, and cluster all the points whose deviation threshold range is within 10 into one cluster, and the largest cluster is the large cluster of the road background map;

b. Cluster all the points of the same value in the largest cluster obtained in step a into one cluster, find the cluster with the largest number, take the longest continuous point, and extract the feature frame; the extraction of the feature frame p _k The model is:

Among them, p _k is the extracted feature frame, p _n is each frame in the largest cluster, n is the number of frames in the largest cluster, i is the starting number of the first frame in the largest cluster, j is the last frame in the largest cluster The number of the frame.

(5) Replace the original frame of the grid with the obtained feature frame, and return to step (4) to start the extraction of the next grid feature frame, until J10 processing is completed, and the real-time background image of the whole road is obtained;

(6) Return to step (1) to start the next round of calculation.

In actual implementation, the grid division method of each frame can be divided into 100 grids, and the number of grids can be determined according to actual needs.

Aiming at the modeling problem of the road background map, the present invention specifically proposes a highly feasible, simple and real-time grid clustering method for obtaining the road background map.

The accompanying drawings are as follows:

FIG. 1 is a segmentation diagram of the first frame of a video.

Fig. 2 is a graph of the H component in the b1 region.

Fig. 3 is a graph of the H component in the b5 region.

Figure 4 is the first frame picture.

Figure 5 is the 900th frame picture.

Figure 6 is a road background synthesis map.

Fig. 7 is a flowchart of the method of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings.

Due to the interference of external sunlight, clouds, and leaves, we cannot get an accurate curve change map. So we find points in the graph that vary within a threshold and group them into clusters. We think this is the cluster of the road background map. Then in this cluster, continue to search for the frame with the largest eigenvalue with steady-state changes, and extract it as the background image of the grid. The grid division is shown in Figure 1, which specifically includes:

(1) Extract the 30-second road video into frames. If the video is 30 frames per second, a total of 900 frames;

(2) Carry out grid interception and division for each frame. We divide each frame into 100 grids, which are represented by two-dimensional matrices of A-J and 1-10, as shown in Figure 1;

(3) Calculate the H component in HSV for each grid of each frame, and count the maximum value of the H component of each grid. The maximum value of the H component in each grid of each frame of 900 frames is plotted as a graph. The fluctuation of the curve indicates that the main color has shifted during this time period, and there are vehicles passing through this area.

(4) Extract feature frames by clustering method:

The extraction formula of feature frame p _k :

p: selected feature frame;

i: in the (4)-b step, the selected starting position;

j: the selected end position in step (4)-b.

a. Starting from grid A1, cluster all the points with deviation thresholds within 10 into one cluster, which can filter out the deviation of H value caused by swinging leaves, sunlight scattering, and cloud cover changes; because the roads in this area It is in the majority when it is free, and it is considered that the largest cluster is the large cluster of the road background map;

b. Continue to process the largest cluster obtained in step a. Cluster all points with the same value in the cluster into one cluster, and find the cluster with the largest number. This allows to find the frame that best represents the road background in the current environment;

c. In the cluster obtained in step b, take the longest continuous point and extract the feature frame. Through this step, it can be further confirmed that at this moment, the roads in this area are in the state where the external influence is the most stable and the roads are in an idle state. We think that this feature frame can represent the real-time background image of this area of the current road, so the original frame can be replaced by this feature frame. As shown in Fig. 2 and Fig. 3, the variation curve of the maximum value of the H component within 30 seconds is drawn.

(5) Replace the original feature frame with the obtained feature frame, and return to step (4) to start the extraction of the feature frame of the next grid until J10 processing is completed;

(6) Return to step (1) to start the next round of calculation, as shown in Figure 7.

From the pictures of the first frame and the 900th frame, it can be found that the picture has a large color shade change, as described in Figure 4, Figure 5, and Figure 6. From the composite image in Figure 6, it can be seen that due to the movement of clouds and changes in light, different grid images have different shades of color. It can be concluded from the experiment that the real-time replacement of the road background image has practical significance for reducing network video traffic.

From the above experimental discussion, it can be known that this method has the following obvious advantages:

1. This method has the advantages of high feasibility and simple coding;

2. The background image obtained by this method has strong real-time performance;

3. The feature frames extracted by this method can respond to many changes such as cloud movement and light changes in time, and have strong anti-interference to the shaking of leaves on both sides of the road.

Claims (1)

1. A method for obtaining the real-time background map of the road, characterized in that, using grid clustering to quickly obtain the method for the background map of the road, at first the road is divided into many grids, and then the statistics of the H components are carried out to these grids, and then Determine the clusters of different video frames according to the H component graph, and finally extract the feature frame from it, including the following processing steps: (1) Extract the 30-second road video into frames, the video is 30 frames per second, a total of 900 frames; (2) Perform grid interception and division for each frame, and each frame is divided into 100 grids, which are represented by two-dimensional matrices of A-J and 1-10; (3) Calculate the H component in HSV for each grid of each frame, and count the maximum value of the H component of each grid; draw the maximum value of the H component in each grid of each frame of 900 frames into a graph ;The fluctuation of the curve indicates that the main color shifted within the 30 seconds, and a vehicle passed the measurement area; (4) Extract feature frames by clustering method: a. Start calculation from grid A1, and cluster all the points whose deviation threshold range is within 10 into one cluster, and the largest cluster is the large cluster of the road background map; b. Cluster all the points of the same value in the largest cluster obtained in step a into one cluster, find the cluster with the largest number, take the longest continuous point, and extract the feature frame; the extraction of the feature frame p _k The model is: Among them, p _k is the extracted feature frame, p _n is each frame in the largest cluster, n is the number of frames in the largest cluster, i is the starting number of the first frame in the largest cluster, j is the last frame in the largest cluster frame number; (5) Replace the original frame of the grid with the obtained feature frame, and return to step (4) to start the extraction of the next grid feature frame, until J10 processing is completed, and the real-time background image of the whole road is obtained; (6) Return to step (1) to start the next round of calculation.