CN103198300A - Parking event detection method based on double layers of backgrounds - Google Patents

Abstract

The invention discloses a parking event detection method based on double layers of backgrounds. The parking event detection method based on the double layers of backgrounds mainly comprises the steps of double-layer background modeling, secondary background replacement, parking detection, and state table updating. According to the parking event detection method based on the double-layer background, an absolute value of the pixel difference of corresponding points of a main background and the second background is used for judging whether a stopping object appears, detection is carried out according to the outline of the stopping object, if the stopping object is a vehicle, a parking state is calibrated and updated, and in addition, the background is replaced by judging whether the foreground in a background model is empty or not. The parking event detection method based on the double layers of backgrounds can well detect a parking event in real time, and is simple and capable of reducing influence of the environment on a detection result by means of the double layers of backgrounds and accurately recording a parking position and the parking state due to the establishment of a parking event state table. The parking event detection method based on the double layers of backgrounds is capable of being used in different occasions such as an expressway, a parking lot, and an urban road due to the setting of detection parameters and interested areas, high in accuracy and good in real-time performance.

Description

Parking event detecting method based on double-deck background

Technical field

The present invention relates to the video detection range, be specifically related to a kind of parking event detecting method based on double-deck background.

Background technology

Along with the swift and violent increase of fast development of national economy and motor vehicles, China's traffic problems are increasingly serious.Motor vehicle is parked, shed traffic jam and the second accident that traffic events such as landslide that thing, traffic hazard stagnation of movement and rugged surroundings cause cause constantly increases, this accident has sudden and contingency, in case take place to cause great personnel and property loss.Parking event detection in the past, mainly by manual monitoring, methods such as telecommunication flow information collection have consumed human and material resources and the financial resources in the traffic control greatly.So, set up particularly important and necessary based on the automatic stopping event detection system of video.

At present develop multiple parking event detecting method based on video in the world, mainly contained based on virtual box pixel, gray-scale statistical based target speed and the parking event detecting method of cutting apart based on single background piece.

Method based on virtual box pixel, gray-scale statistical is obtained motion pixel and static pixel by the background subtraction method, and judges according to the situation of change of pixel in the defined area or gray scale whether vehicle stops; Although the method algorithm is simple, be vulnerable to interference such as extraneous illumination condition, and need the artificial virtual detection zone of setting, it is relatively poor to detect practicality.The parking event detecting method of based target speed need carry out real-time follow-up to vehicle, and scene is demarcated, and needs the movement velocity of real-time calculating target; This algorithm comparatively complexity and rate of false alarm is higher.Cut apart the parking detection method by preserving the different background of three width of cloth based on the piece of single background, relatively determined whether suspicious in twos, suspicious is detected judge the parking event; This method needs that image is carried out piecemeal and calculates, and is not suitable for the complicated location of road conditions, and preserves three width of cloth background images from same background model, and it is not high to detect degree of accuracy.Said method is for vehicle lay-off or sail out of effectively analysis of shortage, and algorithm can not satisfy actual requirement to validity and the practicality of parking event analysis.

Summary of the invention

The object of the present invention is to provide a kind of affected by environment lessly, stop to detect accurately the parking event detecting method based on double-deck background.

Technical solution of the present invention is:

A kind of parking event detecting method based on double-deck background is characterized in that: comprise the following steps: to realize by following steps:

(1) set up two-layer different background---main background and time background, utilize the renewal speed of main background fast, to stopping the more sensitive characteristic of target, and inferior context update is slow, to stopping the slower characteristic of goal response, the difference of two width of cloth figure relatively;

(2) pixel of double-deck background correspondence is made difference and obtained absolute value, obtain static target, this bilayer background error image is carried out binaryzation; Suppress by the HSI shade, and eliminate the shade of target respective pixel point, this bianry image is carried out closed operation, eliminate discontinuous cavity;

(3) according to focal length of camera, height and angle, to target pixel points weighting in the image, target pixel points weights at a distance strengthen, target pixel points weights nearby reduce, obtain object pixel and value after the weighting, when reaching threshold value, parking event counter S adds 1, as S during greater than threshold value, then this bianry image is preserved;

(4) this bianry image is carried out filtering, and the target in the image cut apart with profile detect, detection sensitivity is set, to pixel value greater than the target of the threshold of sensitivity rectangle frame that draws, judge according to length breadth ratio whether this target is vehicle, when target is vehicle, record the diagonal line intersecting point coordinate of this rectangle frame, deposit parking state-event table in;

(5) this pixel is handled and analyzed, judge whether this target is vehicle; If this target is judged as vehicle, trigger the parking affair alarm and also main background present frame is composed to inferior background, proceed diversity ratio; When not having target in the inferior background, keep this two field picture as pure background;

(6) when the prospect in the inferior background model is sky, store current background image as pure background, when having detected when stopping target, compare with main background present frame storage and with this pure background image, continue operation if skip to step (2) when difference is less than threshold value in two two field pictures, when if difference is greater than threshold value in two two field pictures, then replace time background with this main background present frame, skipping to step (2) proceeds, when detecting the parking event again, the comparison object center point coordinate judges whether target sails out of, and the update mode table.

Parking event detecting method based on double-deck background of the present invention with add up based on virtual coil, the parking detection method that based target is followed the tracks of with cut apart the parking detection method based on the piece of single background and compare, affected by environment less, need not image set with image and demarcate, algorithm is simple possible more, reduce the computational complexity that stops and detect, improved the real-time of parking event detection.Suppress and morphologic filtering processing elimination interference by the HIS shade, the feasible detection of stopping is more accurate; Set up state table and upgrade, record parking spot and state accurately.

Description of drawings

The invention will be further described below in conjunction with drawings and Examples.

Fig. 1 is based on the parking detection method process flow diagram of double-deck background.

Embodiment

Provide consistent parking event detecting method based on double-deck background in conjunction with the accompanying drawings and embodiments, this method is by setting up time background---mixed Gaussian background model and main background---RunningAvg background, two-layer background is asked difference, and error image carried out binaryzation, obtain static target or low-speed motion target.Detect by statistics and profile identification to this binary image whether stopped vehicle or legacy are arranged, with this moment RunningAvg background and pure background image contrast, judge whether road is unimpeded, upgrade dead ship condition table and time background, carry out parking event detection next time.Concrete implementation step is as follows:

Step 1 is with the video image I of input _nCarry out the gray processing processing and obtain gray level image I _Ngray, by gray level image I _NgraySet up main background and time background;

Foundation and the renewal of main background, inferior background and pure background:

The foundation of main background model---RunningAvg model, the RunningAvg model is shown below:

$B_{\mathrm{Avg}}(i,j)={\∂}_{\mathrm{avg}}{B}_{n-1}(i,j)+(1-{\∂}_{\mathrm{avg}})I_n(i,j)$

---(i j) is the RunningAvg background model to Bavg;

---B _n(i j) is background value after the n frame upgrades,

---B _N-1(i j) is the background value of n-1 frame,

---I _n(i j) is the gray-scale value of current video frame,

---

Be renewal rate.

The present invention is right Carried out following improvement: ${\∂}_{\mathrm{avg}}={\∂}_{\mathrm{avg}1}{M}_n+{\∂}_{\mathrm{avg}2}(1-M_n)$

—— $M_n=\left\{\begin{array}{cc}0& D_n(i,j)<T\\ 1& D_n(i,j)\&GreaterEqual;T\end{array}\right.,$ M _nBe mode bit;

D _n(i, j)=| I _n(i, j)-I _N-1(i, j) |, (i j) is the consecutive frame residual image to Dn;

Be respectively the variable weighting parameter;

The foundation of inferior background model---predefined several Gauss models of initialization at first carry out initialization to the parameter in the Gauss model, and the parameter that will use after obtaining.Secondly, handle for each pixel in each frame, see whether it mates certain model, if coupling then is included into it in this model, and this model is upgraded according to new pixel value, if do not match, then set up a Gauss model with this pixel, initiation parameter is acted on behalf of model least possible in original model.Select the several most possible models in front model as a setting at last, lay the groundwork for target context extracts.

Mixed Gauss model p (x _N) be pixel probability of occurrence statistical value, be shown below:

$p\left(x_N\right)=\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}{w}_j\mathrm{\&eta;}(x_N;{\mathrm{\&theta;}}_j)$

---w _jBe the weight of k rank Gaussian Background, x _NBe input sample, θ _jBe observed value;

---η (x; θ _k) be k rank Gauss's standardized normal distribution, its expression formula is as follows:

$\mathrm{\&eta;}(x;{\mathrm{\&theta;}}_k)=\mathrm{\&eta;}(x;{\mathrm{\&mu;}}_k,c)=\frac{1}{{\left(2\mathrm{\&pi;}\right)}^{\frac{D}{2}}{\left|{\mathrm{\&Sigma;}}_K\right|}^{\frac{1}{2}}}{e}^{-\frac{1}{2}{(x-{\mathrm{\&mu;}}_k)}^T{\mathrm{\&Sigma;}}_k^{-1}(x-{\mathrm{\&mu;}}_k)}$

---μ _kBe average;

---∑ _k=σ ²I is variance;

The background pixel judgment formula as shown in the formula:

$B_{\mathrm{Gauss}}(i,j)={\arg }_b\min (\underset{j=}{\overset{b}{\mathrm{\&Sigma;}}}{w}_j>T)$

${\hat{w}}_k^{N+1}=(1-a){\hat{w}}_k^N+a\hat{p}\left(w_k\right|x_{N+1})$

---B _GaussBe the background pixel point;

---a is learning rate;

---w _kBe initializes weights,

Be w _kExpectation value;

---T is background threshold;

Mixed Gauss model uses K Gauss model to come the feature of each pixel in the token image, obtain the back at a new two field picture and upgrade mixed Gauss model, with each pixel in the present image and mixed Gauss model coupling, if success then judge that this point is background dot, otherwise be the foreground point.Taking an overall view of whole Gauss model, mainly is to have variance and two parameters of average to determine, to the study of average and variance, takes different study mechanisms, will directly have influence on stability, accuracy and the convergence of model.Because we are the background extracting modelings to moving target, therefore need be to variance in the Gauss model and two parameter real-time update of average.For improving the learning ability of model, improve one's methods different learning rates is adopted in the renewal of average and variance; For improving under busy scene, big and slow motion target detection effect is introduced the concept of weights average, sets up background image and real-time update, in conjunction with weights, weights average and background image pixel is carried out the classification of prospect and background then.

So-called pure background, when inferior background---when the prospect value of mixed Gaussian background model be empty, preservation current images B _{G_clear}As pure background.

Step 2, initialization dead ship condition table, mark are stopped in the scene the whether existing vehicle that stops.

Step 3, the value of double-deck background corresponding pixel points is made difference and is also obtained absolute value, obtain static target or low-speed motion target image D (i, j)=| B _Avg(i, j)-B _Gauss(i, j) |, this bilayer background error image is carried out binaryzation, binary-state threshold T _hSuppress by the HSI shade, and eliminate the shade of target respective pixel point, this image is carried out closed operation, eliminate discontinuous cavity, obtain image D at last _Det(i, j);

Step 4, according to focus of camera and angle to image D _Det(i, j) the white pixel point is weighted, and this example roughly is divided into image 5 parts pixel is weighted.By as far as closely, weights are respectively 2.0,1.6,1.2,1.1 and 1.Pixel after the weighting is added up ought be greater than threshold value T _pThe time, this example T _p=220, counter S adds 1, otherwise S=0; When S more than or equal to 90 the time, then with this binary image D _ObjPreservation is got off;

Step 5 is to image D _ObjCarry out filtering, and the target in the image is cut apart and the profile detection, detection sensitivity T is set _p, to pixel value more than or equal to T _pTarget, detect to determine edge, the upper left corner and the edge, the lower right corner of target with square frame it to be marked by profile, the number of white pixel is removed non-vehicle target in the length breadth ratio by computing block diagram and the frame,

Profile upper left corner point coordinate is (x ₁, y ₁), lower right corner coordinate is (x ₂, y ₂), then long for l=max (| x ₁-x ₂|, | y ₁-y ₂|), wide be w=min (| x ₁-x ₂|, | y ₁-y ₂|).When

During establishment, then this target is vehicle, wherein t ₁, t ₂Length breadth ratio value according to vehicle.When target is vehicle, trigger the parking affair alarm and record the diagonal line intersecting point coordinate of this rectangle frame, deposit parking state-event table in;

Step 6, square frame diagonal line intersecting point coordinate is (x _I1, y _I1) wherein

And storing coordinate information, when detecting the stagnation of movement target again and be vehicle, obtain next group vehicle limit block diagonal line intersecting point coordinate (x _J1, y _J1), when max (| x _I1-x _J1|, | y _I1-y _J1|)≤Tc then thinks has vehicle to sail out of, and eliminates the vehicle coordinate of storing in the state table, upgrades to stop vehicle fleet size, and namely automobile storage subtracts 1 in value.When max (| x _I1-x _J1|, | y _I1-y _J1|) T _cThe time, then thinking has vehicle to sail into, increases vehicle coordinate in the state table, upgrades to stop vehicle fleet size, and namely automobile storage adds 1 in value, wherein T _cBe the side-play amount threshold value, and

W is overall width;

With current RunnineAvg background storage and with pure Gaussian Background image B _{G_clear}Compare, if difference is less than threshold value T in two two field pictures _d, skip to step 3 and continue to carry out, if difference is more than or equal to threshold value T in two two field pictures _d, then replace Gaussian Background with this RunnineAvg background, and skip to step 3 continuation execution;

Main context parameter setting, the renewal speed parameter

Inferior context parameter setting, Gaussian distribution weight sum threshold value T=0.7, background threshold T=2.5, learning rate

Initial weight w _k=0.05, primary standard difference ∑ _k=30;

In the step 3, threshold value T _h=25; In the step 5, detection sensitivity T _p=200.

Claims (1)

1. parking event detecting method based on double-deck background is characterized in that: comprise the following steps: to realize by following steps:

(1) set up two-layer different background---main background and time background, utilize the renewal speed of main background fast, to stopping the more sensitive characteristic of target, and inferior context update is slow, to stopping the slower characteristic of goal response, the difference of two width of cloth figure relatively;

(2) pixel of double-deck background correspondence is made difference and obtained absolute value, obtain static target, this bilayer background error image is carried out binaryzation; Suppress by the HSI shade, and eliminate the shade of target respective pixel point, this bianry image is carried out closed operation, eliminate discontinuous cavity;

(3) according to focal length of camera, height and angle, to target pixel points weighting in the image, target pixel points weights at a distance strengthen, target pixel points weights nearby reduce, obtain object pixel and value after the weighting, when reaching threshold value, parking event counter S adds 1, as S during greater than threshold value, then this bianry image is preserved;

(4) this bianry image is carried out filtering, and the target in the image cut apart with profile detect, detection sensitivity is set, to pixel value greater than the target of the threshold of sensitivity rectangle frame that draws, judge according to length breadth ratio whether this target is vehicle, when target is vehicle, record the diagonal line intersecting point coordinate of this rectangle frame, deposit parking state-event table in;

(5) this pixel is handled and analyzed, judge whether this target is vehicle; If this target is judged as vehicle, trigger the parking affair alarm and also main background present frame is composed to inferior background, proceed diversity ratio; When not having target in the inferior background, keep this two field picture as pure background;

(6) when the prospect in the inferior background model is sky, store current background image as pure background, when having detected when stopping target, compare with main background present frame storage and with this pure background image, continue operation if skip to step (2) when difference is less than threshold value in two two field pictures, when if difference is greater than threshold value in two two field pictures, then replace time background with this main background present frame, skipping to step (2) proceeds, when detecting the parking event again, the comparison object center point coordinate judges whether target sails out of, and the update mode table.