CN102970517A

CN102970517A - Holder lens autonomous control method based on abnormal condition identification

Info

Publication number: CN102970517A
Application number: CN2012104924315A
Authority: CN
Inventors: 江佳峻; 张成亮; 刘威
Original assignee: Sichuan Changhong Electric Co Ltd
Current assignee: Sichuan Changhong Electric Co Ltd
Priority date: 2012-11-28
Filing date: 2012-11-28
Publication date: 2013-03-13
Anticipated expiration: 2032-11-28
Also published as: CN102970517B

Abstract

The invention relates to video monitoring technology, in particular to a method for autonomous control and monitoring of pan-tilt lens based on abnormal scene recognition with intelligent image analysis. The main steps of the method of the present invention are: the range of rotation of the pan-tilt lens is divided into several blocks, and the feature vector of each block is obtained; the current feature vector of each block is obtained and performed with the corresponding normal feature vector Compare and obtain covariance; judge whether there is any abnormality in the video image of the current block, lock the abnormal target, and the pan-tilt lens will automatically track the abnormal target and drive the pan-tilt lens to rotate in time through prediction. The beneficial effect of the present invention is that the abnormal scene is automatically detected through the video image analysis technology, and its trajectory is analyzed and predicted, and then the pan-tilt lens automatically controls its rotation to track the abnormal target in real time, so that the monitoring personnel can perform monitoring tasks more efficiently and quickly . The invention is especially suitable for a pan-tilt lens monitoring system.

Description

The autonomous control method of platform-lens of anomaly-based sight identification

Technical field

The present invention relates to Video Supervision Technique, particularly relate to a kind of platform-lens of identifying with the anomaly-based sight of image intelligent analysis and monitor method from master control.

Background technology

Along with the development of computer soft or hard technology, be widely used in every field based on the video monitoring system of multimedia technology.In present many markets, bank and the high-grade residential quarter platform-lens supervisory control system has been installed all, some industrial and mining enterprises, bank vault and military depot also wish to utilize the platform-lens supervisory control system to realize intellectuality, automation unattended operation.Therefore the platform-lens supervisory control system of intelligence has very large market.Present platform-lens supervisory control system is monitor staff's persistent surveillance screen normally, and the target that enters guarded region is finished motion tracking to target by the operation keyboard rocking bar.And when target moves away from video camera, also need to control manually the camera lens zoom, target is carried out feature amplify.In manually controlling the real-time tracking and zoom process of platform-lens realization to target, also may there be artificial tracking error.Therefore, the present technology of utilizing platform-lens to monitor exists degree of intelligence lower, the situation of error occurs easily.

Summary of the invention

This reality invention technical problem to be solved is, is exactly for the lower problem of present platform-lens monitoring intelligent degree, proposes a kind of autonomous control method of platform-lens of anomaly-based sight identification.

The present invention solves the problems of the technologies described above the technical scheme that adopts: the autonomous control method of platform-lens of anomaly-based sight identification, it is characterized in that, and may further comprise the steps:

A. the slewing area with platform-lens is divided into several blocks, and the video image collection of each block under normal sight is set to an independent sample collection, with the gray scale of each pixel in each sample image characteristic vector as the independent sample collection;

B. gather the video image of current each block by platform-lens, obtain current block video image characteristic vector and compare with the characteristic vector of corresponding independent sample collection, obtain covariance;

C. judge according to covariance whether the video image of current block is in the early warning interval, if, then enter steps d, if not, then get back to step b;

D. obtain the video image of current block and the gray scale difference value of corresponding independent sample collection, gray scale difference value is locked as abnormal object greater than the zone of set point continuously, platform-lens carries out automatic tracing to abnormal object;

E. the video image of abnormal object block appears in continuous acquisition, and the movement locus of abnormal object is predicted, drives platform-lens according to predicting the outcome and in time rotates to reduce lag time.

Concrete, described independent sample collection comprises 100 video images at least.

Concrete, among the step a with the gray scale of each pixel in each sample image as the concrete steps of the characteristic vector of independent sample collection be:

A1. with the gray scale of each each pixel of sample image as characteristic vector, input neural network is trained;

A2. the output of neural net is vectorial at the gray feature of normal contextual model as this block.

Concrete, described neural net is the RBF radial base neural net.

Beneficial effect of the present invention is, when being implemented in Real Time Monitoring, by the unusual sight of video image analysis technology Auto-Sensing, and its movement locus of analyses and prediction, then it rotates with the real-time tracking abnormal object platform-lens from master control, thereby makes the monitor staff more carry out monitor task in efficient quick ground.

Description of drawings

Fig. 1 is method flow diagram of the present invention.

Embodiment

Be described further below in conjunction with the method for accompanying drawing to invention:

As shown in Figure 1, the present invention proposes a kind of autonomous control method of platform-lens of anomaly-based sight identification, key step is: at first the slewing area with platform-lens is divided into several blocks, the video image collection of each block under normal sight is set to an independent sample collection, with the gray scale of each pixel in each sample image characteristic vector as the independent sample collection, generally in order to ensure monitoring range, can divide the block more than 3; Gather the video image of current each block by platform-lens, obtain current block video image characteristic vector and compare with the characteristic vector of corresponding independent sample collection, obtain covariance; Whether the video image of judging current block according to covariance is in the early warning interval, usually the early warning interval can be arranged on [30, + 30] between, if, then continue to obtain the video image of current block and the gray scale difference value of corresponding independent sample collection, with gray scale difference value continuously greater than set point regional as, platform-lens carries out automatic tracing to abnormal object, if not, then continue to gather the video image of each block; Obtain the video image of current abnormal object block and the gray scale difference value of corresponding independent sample collection, gray scale difference value is locked as abnormal object greater than the zone of set point continuously, here also blocked abnormal object can be placed the centre position of display interface, with better prompting observer, and usually set point can be set as between the 40-80, platform-lens carries out automatic tracing to abnormal object; The video image of abnormal object block appears in continuous acquisition, and the movement locus of abnormal object predicted, drive platform-lens according to predicting the outcome and in time rotate to reduce lag time, here used prediction algorithm can be recurrent least square method, also can keep all the time the centre position that abnormal object is placed display interface.

A kind of concrete gray scale with each pixel in each sample image as the concrete steps of the characteristic vector of independent sample collection is: at first with the gray scale of each each pixel of sample image as characteristic vector, input neural network is trained; Then the output of neural net is vectorial at the gray feature of normal contextual model as this block.

Concrete, described neural net is the RBF radial base neural net.

Claims

1. The method for autonomously controlling the pan-tilt lens based on abnormal situation recognition, is characterized in that, comprises the following steps:

a. Divide the rotation range of the pan/tilt lens into several blocks, set the video image set of each block under normal conditions as an independent sample set, and use the grayscale of each pixel in each sample image as an independent sample set. The feature vector of the sample set;

b. Collect the video image of each current block through the pan-tilt lens, obtain the feature vector of the video image of the current block and compare it with the feature vector of the corresponding independent sample set to obtain the covariance;

c. Judging whether the video image of the current block is in the early warning interval according to the covariance, if so, then enter step d, if not, then return to step b;

d. Obtain the grayscale difference between the video image of the current block and the corresponding independent sample set, and lock the area where the grayscale difference is continuously greater than the set value as an abnormal target, and the PTZ lens will automatically track the abnormal target;

e. Continuously collect video images of abnormal target blocks, and predict the trajectory of the abnormal target, and drive the pan-tilt lens to rotate in time according to the prediction results to reduce the lag time.

2. The method for autonomously controlling the pan/tilt lens based on abnormal scene recognition according to claim 1, wherein the independent sample set includes at least 100 video images.

3. the PTZ lens autonomous control method based on abnormal scene recognition according to claim 1, is characterized in that, in the step a, the grayscale of each pixel point in each sample image is used as the concrete step of the feature vector of independent sample set for:

a1. Use the grayscale of each pixel of each sample image as a feature vector, and input it into the neural network for training;

a2. The output of the neural network is used as the gray feature vector of the block in the normal scene mode.

4. The method for autonomously controlling the pan-tilt lens based on abnormal scene recognition according to claim 3, wherein the neural network is an RBF radial basis neural network.