CN108765451A - A kind of movement of traffic object detection method of adaptive RTS threshold adjustment - Google Patents

Abstract

The present invention discloses a kind of movement of traffic object detection method of adaptive RTS threshold adjustment, solves the problems, such as that movement of traffic target detection is easy to be influenced by outer scene variation.The present invention uses median method to extract background image that is reliable, stablizing from sequence of video images first, then current frame image and the background image of extraction are made into difference, set determining threshold value, obtain error image, certain post processing of image is finally carried out to error image, including the morphologic processing such as burn into expansion and opening and closing, you can obtain required movement of traffic target.The present invention can effectively improve the accuracy of movement of traffic target detection, have good real-time and robustness.

Description

A kind of movement of traffic object detection method of adaptive RTS threshold adjustment

Technical field

The present invention relates to a kind of movement of traffic object detection methods of adaptive RTS threshold adjustment, belong to Activity recognition, target The interleaving techniques fields such as detection.

Background technology

The detection of moving target is a step crucial in computer vision technique and target identification and processing in image sequence, In recent years, being used to detect the moving target in video image there has been proposed many methods, general way is prominent movement Target object eliminates background image, mainly there is optical flow method, frame differential method and background subtraction etc., and different methods is each There are advantage and disadvantage, is suitable for different situations.

For background subtraction compared with other object detection methods, biggest advantage is opposite in the background image due to extraction Stablize, influence of the moving target to background image itself can be reduced, more accurately extracts the region of moving target, but background The obtained background image of calculus of finite differences can change with the change of time, caused by the external environment conditions such as illumination and weather Scene changes are also more sensitive, thus will appear pseudo-motion target point, can thus influence the effect of target detection.Background difference Method is easily achieved, can be more accurate in the case where background is stablized detect moving target object, disadvantage be exactly be easy by The influence of the external conditions such as ambient light variation, this just needs constantly to update background image, to adapt to background image Continuous variation.

Invention content

Technical problem：The context update detected the technical problem to be solved by the present invention is to operational objective in traffic video is asked Topic, effectively to improve to movement of traffic target detection accuracy.

Technical solution：A kind of movement of traffic object detection method of adaptive RTS threshold adjustment of the present invention includes following step Suddenly：

Step 1) inputs traffic video, by the kth frame image F of movement of traffic target to be detected in the video_kAs current Frame；

Step 2) obtains initial background image B using median method_k, to each pixel (x, y), the acquisition of background uses Following formula：

The B (x, y) is the gray value of pixel in initial background image, F_i(x, y) is the i-th frame image in image sequence The gray value of pixel, N are that certain time period samples to obtain the frame number of image；

Step 3) is updated background image：First with the relevance between current frame image and background image, Current frame image is divided into m*m blocks respectively with former background image, and then newer object is current frame image and former background image In each fritter, the update between each block is independent, does not influence, is updated to each block diagram picture between each other, finally right Each updated block diagram picture is overlapped, that is, obtains new background image, more new formula is as follows：

B_{I, k}(x, y)=(1-a) B_{I, k_1}(x, y)+af_{I, k}(x, y) (1≤i≤m²)

The i indicates the i-th fritter in image, 1≤i≤m², k and k-1 indicate present image frame number and previous image Frame number, f_{I, k}(x, y) indicates i-th piece of current frame image, B_{I, k-1}(x, y) indicates i-th of the former background image before context update Block, B_{I, k}(x, y indicate i-th piece of new background image, B_k(x, y) indicates that the new background image after context update, a indicate update system Number, value range are 0~1；

Step 4) carries out background difference and binary conversion treatment, and processing formula is as follows：

D_k(x, y)=| F_k(x, y)-B_k(x, y) |

The T is the color difference threshold of setting, formula D_k(x, y)=| F_k(x, y)-B_k(x, y) | it calculates by pretreatment Current frame image F_kWith background image B_kDifference, by formulaTo differential chart As D_kCarry out binary conversion treatment；

Step 5) is to binary image R_kThe corrosion in mathematical morphology and dilation operation are carried out, movement of traffic mesh is obtained Mark, formula are as follows：

A is corroded with B：

A is expanded with B：

Wherein, Φ indicates empty set.

It is preferred that the specific method is as follows for the step 3)：

Step 3.1) will be divided into 4 × 4 block per frame image；

Step 3.2) is formula B to each piece_{I, k}(x, y)=(1-a) B_{I, k-1}(x, y)+af_{I, k}(x, y) (1≤i≤m²) Update is calculated, a takes 0.5；

Step 3.3) repeats step (2), and every piece until background image is all updated；

Step 3.4) utilizes formulaNew background is obtained to all block image additions.

It is preferred that color difference threshold T takes 95 in step 4).

Advantageous effect：The present invention has the following technical effects using above technical scheme is compared with the prior art：

The present invention is based on background subtractions to be constantly updated background image, be made background using adaptive context update algorithm Influence of the outer scene variation to movement of traffic target detection effect can be reduced constantly close to perfect condition.

Description of the drawings

Fig. 1 is the target vehicle detection method flow diagram of adaptive RTS threshold adjustment.

Specific implementation mode

The some embodiments of attached drawing of the present invention are described in more detail below：

With reference to the accompanying drawings 1, the specific embodiment of the invention is：Input has the traffic video of operational objective, will be waited in the video Detect the kth frame image F of movement of traffic target_kAs present frame；Input video obtains current frame image F_k；It is obtained using median method Take initial background image B_k, to each pixel (x, y), useObtain background.Having During body is implemented, B (x, y) is the gray value of pixel in initial background image, F_i(x, y) is the i-th frame image slices in image sequence The gray value of vegetarian refreshments, N are that certain time period samples to obtain the frame number of image.

The present invention proposes a kind of improved background update method, is carried out more to background image using the thought of piecemeal Newly.Specific implementation process is：When being updated to preceding background image, make full use of between current frame image and background image Current frame image and former background image are divided into bulk by relevance, and then newer object is current frame image and former background Each fritter in image, and the update between each block is independent, is not influenced between each other, is carried out to each block diagram picture Corresponding update, is finally overlapped each updated block diagram picture, you can obtain new background image, reach to Background As updating in real time and accurately, more new formula is：B_{I, k}(x, y)=(1-a) B_{I, k-1}(x, y)+af_{I, k}(x, y) (1≤i≤m²),The i indicates that the i-th fritter in image, k and k-1 indicate present image frame number and previous figure The frame number of picture, f_{I, k}(x, y) indicates i-th piece of current frame image, B_{I, k-1}(x, y) indicates the former background image before context update I-th piece, B_{I, k}(x, y) indicates i-th piece of new background image, B_k(x, y) indicates that the new background image after context update, a indicate more New coefficient, value range are 0~1.

The specific implementation step for updating background image is as follows：

(1) block of m × m will be divided into per frame image.The number of piecemeal determines that piecemeal is more, as a result will by testing It is more accurate, but calculation amount also can be bigger, and it is appropriate to be generally divided into 4 × 4；

(2) each piece of calculating for doing formula (2) is updated；

(3) step (2) is repeated, every piece until background image is all updated；

(4) formula (3) is utilized, new background is obtained to all block image additions；

Above-mentioned steps are completed, background difference is being carried out below and binary conversion treatment, formula is as follows：

D_k(x, y)=| F_k(x, y)-B_k(x, y) |

In specific implementation, above-mentioned T is the threshold value of setting, and T takes 95, and formula (4) first is calculated by pretreated Current frame image F_kWith background image B_kDifference, then by formula (5) to error image D_kCarry out binary conversion treatment；

In specific implementation, below to binary image R_kThe corrosion in mathematical morphology and dilation operation are carried out, when A is used B corrodes：When A is expanded with B：Both traffic can be obtained Moving target.

Claims (3)

1. a kind of movement of traffic object detection method of adaptive RTS threshold adjustment, which is characterized in that include the following steps：

Step 1) inputs traffic video, by the kth frame image F of movement of traffic target to be detected in the video_kAs present frame；

Step 2) obtains initial background image B using median method_k, to each pixel (x, y), the acquisition of background uses following formula：

The B (x, y) is the gray value of pixel in initial background image, F_i(x, y) is the i-th frame image pixel in image sequence The gray value of point, N is that certain time period samples to obtain the frame number of image；

Step 3) is updated background image：First with the relevance between current frame image and background image, current Frame image is divided into m*m blocks respectively with former background image, and then newer object is in current frame image and former background image Each fritter, the update between each block is independent, does not influence, is updated to each block diagram picture, finally to each between each other Updated block diagram picture is overlapped, that is, obtains new background image, and more new formula is as follows：

B_{I, k}(x, y)=(1-a) B_{I, k-1}(x, y)+af_{I, k}(x, y) (1≤i≤m²)

The i indicates the i-th fritter in image, 1≤i≤m², the frame number of k and k-1 expression present image frame number and previous image, f_{I, k}(x, y) indicates i-th piece of current frame image, B_{I, k-1}(x, y) indicates i-th piece of the former background image before context update, B_{I, k} (x, y) indicates i-th piece of new background image, B_k(x, y) indicates that the new background image after context update, a indicate update coefficient, Value range is 0~1；

Step 4) carries out background difference and binary conversion treatment, and processing formula is as follows：

D_k(x, y)=| F_k(x, y)-B_k(x, y) |

The T is the color difference threshold of setting, formula D_k(x, y)=| F_k(x, y)-B_k(x, y) | it calculates and works as by pretreated Prior image frame F_kWith background image B_kDifference, by formulaTo error image D_k Carry out binary conversion treatment；

Step 5) is to binary image R_kThe corrosion in mathematical morphology and dilation operation are carried out, movement of traffic target, formula are obtained It is as follows：

A is corroded with B：

A is expanded with B：

Wherein, Φ indicates empty set.

2. a kind of movement of traffic object detection method of adaptive RTS threshold adjustment according to claim 1, which is characterized in that The specific method is as follows for the step 3)：

Step 3.1) will be divided into 4 × 4 block per frame image；

Step 3.2) is formula B to each piece_{I, k}(x, y)=(1-a) B_{I, k-1}(x, y)+af_{I, k}(x, y) (1≤i≤m²) calculating Update, a take 0.5；

Step 3.3) repeats step (2), and every piece until background image is all updated；

Step 3.4) utilizes formulaNew background is obtained to all block image additions.

3. a kind of movement of traffic object detection method of adaptive RTS threshold adjustment according to claim 1, which is characterized in that Color difference threshold T takes 95 in step 4).