Based on image processing in railway video monitoring algorithm design (1)

0 introduction

In recent years, many relating to life and property safety industry security monitoring is more and more people's attention.

Railway safety monitoring along the railways and the focus of the project, along with the constant speed of the train, safety issues cannot be neglected, how to ensure the safety and the safety of pedestrians is directly related to the development of the railway industry.

This design of railway video monitoring method is through a video image processing realization for pedestrian crossing railway lines of the monitor.

When the monitored area after a goal can be auto-alarm and store the corresponding fault photographs afterwards work provides reliable basis. Algorithm design biggest characteristics presented in image processing for core security monitoring design, image processing can automatically determine any dangerous situations and automatic alarm, especially in the monitoring personnel are not able to play a significant role.

1 video monitoring system background extraction

Background-image is the field in the image without moving targets, even as the venue has a moving target, background processing algorithms through technology and get dropped goal without moving target background image.

The current general background extraction methods have the histogram statistics, statistical method, the values in a multiple-frame image averaging and continuous frame, etc.

These four methods have separate faults.

Statistics histogram problem is with the statistics frame count increases, the background image effects and not obvious; statistical median method of statistics histogram also desserts, in addition to the algorithms implemented for compute larger, take up computer memory larger, slower processing; multiple-frame image average method to get a background image from the target activity of relatively large, with increases in the average number of frames, noise elimination will be improved; continuous frame static background image can not be directly obtained, the key is how targeted campaigns to get good background image, because the algorithm did not frame difference itself further processing, the problem is easy to put a texture similar to the prospects of the overlap area mistaken for background.

Considering the above four background extraction methods, this article uses the combination of multiple-frame image averaging and continuous frame both methods for the extraction of the background video sequences.

Taking into account the diversity of moving objects, targets after field changes arising for an extended period of time may be ignored. Multiple-frame image averaging is a moving target as noise, with cumulative average of targets for the Elimination of noise, use run-time image sequences for average and field of view background image. Continuous frame difference algorithm is passed the current frame of the image and the previous frame image difference found movement area, sports area of the background remains unchanged, and the non-movement area background is used to update the current frame, after a period of iteration can extract out the background. The combination of the two methods is a good way to compensate for their weaknesses, able to adapt to the surrounding environment along the railway, to present a moving target of background can extract better quality background images. Specific process is as follows:

Only three frames, the first from a video sequence in casual extract three frames, as Figure 1 (a) «Figure 1 (c) below.

Then extracted by three images are converted to grayscale map and find the average of their grayscale image to grayscale average image as continuous frame of the first frame image (that is, the raw image). Extract from the original image and the image of three frames to do continuous frame, the background image is grayscale image as in Figure 1 (d) below.

2 field of view in the area of hazardous areas

From the background image is grayscale, conduct a series of image processing, will get a calibration of danger zone of the binary image.

First of all on the background to gray, homogenizing is image enhancement, image binarization, regional flags, remove images rough and image expansion.

Image binarization result shown in Figure 2 (a) below, the final result as in Figure 2 (b) below.

From the comparison of two images, the image can be seen through a series of basic calibration after processing, the danger zone. For the figure 2 (b), the white area in the diagram as a twin-track railway is set within the danger zone. Use this method to get the results of the image on the lighting conditions sufficient to get better results, do not apply to the lighting conditions and dangerous regional calibration at night.

[1] [2] [3]