CN106203302A - Pedestrian detection that view-based access control model and wireless aware combine and statistical method - Google Patents

Abstract

The invention mainly relates to view-based access control model and pedestrian detection that wireless aware combines and statistical method, comprise the following steps: (1), monitoring image collection；(2), Image Information Processing: obtain the coordinate of the scope box of every frame line people's quantity and picture position, pedestrian place；(3), wireless signal acquiring: use libtins to obtain claim frame data, the mac address of analyzing device and signal intensity；(4), wireless signal processes: obtain the wireless exploration equipment spatial model to the spacing being calibrated equipment；(5), target is slightly mated: the detection target of camera head mated with the detection target of wireless exploration equipment；(6), target essence coupling: assess the matching degree of the detection target of camera head and the detection target of wireless exploration equipment according to positional information and motion vector；(7), conclude the data detecting region internal object, remove repetition target, the detection target of integration camera head and No. ID of the detection target of wireless exploration equipment, and unique ID number and real-time position information are stored in chained list.

Description

Pedestrian detection that view-based access control model and wireless aware combine and statistical method

Technical field

The invention mainly relates to a kind of detection and statistical method, tie mutually with wireless aware more particularly, to view-based access control model The pedestrian detection closed and statistical method.

Background technology

The pedestrian detection method using view-based access control model belongs to the main stream approach of object detection field from present's view, from 2005 The proposition of year Hog characteristic detection method, the application of machine vision also becomes more and more extensive, has emerged large quantities of outstanding calculation Method and leading edge operation person, wherein object detection methods based on parts are the main flow algorithms that current accuracy of detection is higher, and it has all Many advantages:

1, having various features model, it is divided into master cast and submodel, and utilizes spring deformation principle to calculate detection mesh Target final score, therefore shooting angle and deformation to human body have preferable robustness.

2, use Hog pyramid that the detection image of multiple yardstick is successively mated, be greatly improved various sizes of The recall rate of pedestrian's pixel.

3, insensitive to the circumstance of occlusion occurred in real process, actual effect can receive preferable effect.

Along with improving constantly of living standards of the people, more people can carry such as mobile phone, panel computer when trip Deng mobile terminal device, the collection of the wireless signal that mobile terminal device is sent therefore can be used to carry out indirect detection pedestrian, Complementation can be carried out with vision-based detection, compensate for the vision dead zone of video camera, and reduce because processing target in image and deliberately blocking, The missing inspection that large scale deformation and uneven illumination cause and detection Problem of Failure.

Summary of the invention

Present invention generally provides a kind of view-based access control model and pedestrian detection that wireless aware combines and statistical method, it is right to use The collection of the wireless signal that mobile terminal device sends carrys out indirect detection pedestrian, can carry out complementation with vision-based detection, compensate for The vision dead zone of video camera, and reduce because processing target in image and deliberately blocking, large scale deformation and uneven illumination cause Missing inspection and detection Problem of Failure.

For solving above-mentioned technical problem, the pedestrian detection that view-based access control model of the present invention and wireless aware combine and statistics side Method, it is characterised in that comprise the following steps:

(1), monitoring image collection: utilize camera head to gather video monitoring image, the image obtained is carried out size change Change；

(2), Image Information Processing: the imagery exploitation DPM method obtaining step (1) extracts the ROI in image, carries out color Color passage conversion, calculates feature pyramid, coupling master cast and submodel parts, calculating unit score, screens and demarcate Divide and exceed the target area setting threshold value, it is thus achieved that the coordinate of the scope box of every frame line people's quantity and picture position, pedestrian place；

(3), wireless signal acquiring: what pedestrian was carried by the wireless exploration equipment that utilization is arranged in detection region has nothing The calibration facility of line signal transmitting and receiving function makes requests on frame search, uses libtins to obtain claim frame data, analyzing device Mac address and signal intensity；

(4), wireless signal process: according to signal attenuation formula Los=32.44+20lgd+20lgf,

In formula, Los is propagation loss, and unit is dB；D is distance, and unit is Km；F is operating frequency, and unit is MHz；

Signal intensity in detection region is modeled, it is thus achieved that wireless exploration equipment is to the sky of the spacing being calibrated equipment Between model；

(5), target is slightly mated: sets up the camera head space coordinates to the spacing of pedestrian, divides monitoring image and exist Corresponding domain in space, mates the detection target of camera head with the detection target of wireless exploration equipment, it is thus achieved that same Target sequence number corresponding relation in two data fields；

(6), target essence coupling: according to the movement relation of front and back's frame target, detection target and the nothing to camera head simultaneously The detection target of line detecting devices carries out Kalman in the position of subsequent time and estimates, it is thus achieved that its moving displacement vector, according to position The detection target of confidence breath and motion vector assessment camera head and the matching degree of the detection target of wireless exploration equipment, arranged Whether two targets are the matching value of same target；

(7), conclude the data detecting region internal object, remove repetition target, integrate detection target and the nothing of camera head No. ID of the detection target of line detecting devices, and unique ID number and real-time position information are stored in chained list.

As the further optimization of this programme, pedestrian detection that view-based access control model of the present invention and wireless aware combine and statistics Transforming image dimension in step described in method (1) is suitable for VGA-QVGA.

As the further optimization of this programme, pedestrian detection that view-based access control model of the present invention and wireless aware combine and statistics Color channel described in method is transformed to be converted to RGB triple channel information single-channel data, the master cast number in described step (2) Amount is 2, and submodel number of components is 5, and the feature pyramid iteration number of plies is 10-48 layer, score threshold range set value be [- 1.8f,1.0f]。

As the further optimization of this programme, pedestrian detection that view-based access control model of the present invention and wireless aware combine and statistics Calibration facility in step described in method (3) is wireless network card or wireless router.

As the further optimization of this programme, pedestrian detection that view-based access control model of the present invention and wireless aware combine and statistics The number of calibration facility described in method is more than 2.

As the further optimization of this programme, pedestrian detection that view-based access control model of the present invention and wireless aware combine and statistics In step described in method (4), spatial modeling should model with wireless exploration equipment position, image division angle [45 °, 180°]。

As the further optimization of this programme, pedestrian detection that view-based access control model of the present invention and wireless aware combine and statistics The allowable error of the matching value in step described in method (6) not should be greater than 3m.

Pedestrian detection that view-based access control model of the present invention and wireless aware combine and statistical method have the beneficial effect that use is right The collection of the wireless signal that mobile terminal device sends carrys out indirect detection pedestrian, can carry out complementation with vision-based detection, compensate for The vision dead zone of video camera, and reduce because processing target in image and deliberately blocking, large scale deformation and uneven illumination cause Missing inspection and detection Problem of Failure.

Accompanying drawing explanation

The present invention will be further described in detail with specific implementation method below in conjunction with the accompanying drawings.

Fig. 1 is view-based access control model of the present invention and wireless aware combines pedestrian detection and the flow chart of statistical method.

Fig. 2 is vision-based detection partial process view in the present invention.

Fig. 3 is signal detection partial process view in the present invention.

Detailed description of the invention

Illustrate what patent of the present invention, view-based access control model described in patent of the present invention and wireless aware combined in conjunction with Fig. 1,2,3 Pedestrian detection and statistical method, comprise the following steps:

(1), monitoring image collection: utilize camera head to gather video monitoring image, use resize function in Opencv The image obtained is carried out size change over, and transforming image dimension is suitable for VGA-QVGA；

(2), Image Information Processing: the imagery exploitation DPM method obtaining step (1) extracts the ROI in image, utilizes and turns Changing formula Grey=0.03*red+0.59*green+0.11*blue and carry out color channel conversion, color channel is transformed to RGB Triple channel information is converted to single-channel data, and in conversion formula, grey represents that gray value, red represent the value of R passage, green table Showing the value of G passage, blue represents the value of channel B；

Calculate feature pyramid and first carry out channel characteristics scaling:

Law between channel characteristics and each yardstick:

C_S=Ω (R (I, s))

Wherein Ω is any translation invariant Feature Mapping function, and I is given input picture, characteristics of image C=Ω (I), C is called channel characteristics, C_SFor the channel characteristics of S layer, R is sampling function, and first zoomed image I is to yardstick s, then calculates I_SFeature,

Then quick approximate formula is used

$C_S^{\mathrm{\Ω}}\≈R(C^{\mathrm{\Ω}},s)\×s^{-{\mathrm{\λ}}_{\mathrm{\Ω}}}$

Wherein s take [1,1/2,1/4 ...], calculate feature pyramid, coupling master cast and submodel parts；

Calculating unit score:

$score(x_0,y_0,l_0)=R_{0,l_0}(x_0,y_0)+{\mathrm{\Σ}}_{i=1}^n{D}_{i,,l_0-\mathrm{\λ}}\left(2\right(x_0,y_0)+v_i)+b$

Score represents target PTS, and it comprises three variable l₀, x₀, y₀,(x₀, y₀) it is master cast wave filter (Root Filter) left summit is at the coordinate of master cast characteristic pattern (Root Feature Map), l₀For the number of plies at master cast place, its R_0,l0(x₀,y₀) it is the score of master cast,Score for submodel.Parameter b be in order to main mould The compensation that type directly aligns and arranges, wherein (x₀, y₀) it is that the left summit of master cast wave filter (Root Filter) is at master cast The coordinate of characteristic pattern (Root Feature Map), 2 (x₀, y₀)+v_iIt is mapped to for i-th parts wave filter (Part Filter) Coordinate in partial model characteristic pattern (Part Feature Map), owing to the segmentation of partial model characteristic pattern is master cast feature The twice of figure, it is therefore desirable to be multiplied by twice.v_iIllustrate the skew on the left summit relative to master cast wave filter.

Wherein, the score of parts wave filter is as follows:

D_{I, l}(x, y)=max_{Dx, dv}(R_i,_l(x+dx, y+dy)-d_i·Φ_d(dx, dy))

D_i,l(x, y) represents the score of parts wave filter, and i is unit number, d_iFor offset vector (d_z,d_v,d_z ²,d_v ²), Ф_d (d_x,d_y) it is the Cost weights offset, such as Ф_d(d_x,d_y)=(0,0,1,1) then d_iФ_d(d_x,d_y) it is most common Euclidean Distance, this step is referred to as range conversion.

At parts wave filter ideal position, (x, y), i.e. in the anchor point position certain limit of parts, finds a comprehensive matching The position optimum with deformation.

Set threshold value m and the target score in object detection results vector detetion is circulated judgement, if less than m, From detetion, remove this target, thus screening and demarcation score exceed the target area setting threshold value.

The size of the detection vector detected is read out, thus obtains the number of target, the coordinate to target Determine use non-maxima suppression method, coordinates of targets generally have multiple parts detection box determine, therefore box be set as The matrix of one m*n, wherein m is the number of box, and front the 4 of n are classified as the coordinate of each box, and form is (x1, y1, x2, y2), obtains Obtain the coordinate of the scope box of every frame line people's quantity and picture position, pedestrian place；Master cast quantity is 2, and submodel number of components is 5, the feature pyramid iteration number of plies is 10-48 layer, and score threshold range set value is [-1.8f, 1.0f].

(3), wireless signal acquiring: what pedestrian was carried by the wireless exploration equipment that utilization is arranged in detection region has nothing The calibration facility of line signal transmitting and receiving function makes requests on frame search, uses libtins to obtain claim frame data, analyzing device Mac address and signal intensity；Calibration facility is wireless network card or wireless router, and the number of calibration facility is more than 2.

Obtain the title of native network equipment, be saved in set deviceArr

Set<std::string>iface_set=Utils::network_interfaces ()；

Arranging network interface card is monitoring pattern

Sniffer sniffer(iface,Sniffer::PROMISC,"",true)；

The analytical function that should run when often capturing a signal is set

sniffer.sniff_loop(make_sniffer_handler(this,&ProbeSniffer:: callback))；

ProbeRequest frame is looked in the data captured

If capturing probe frame

Probe.to_ds ()==0&&probe.from_ds ()==0

Then extract radiotap frame,

This frame storage signal intensity

Const RadioTap&radio=pdu.rfind_pdu<RadioTap>()；

Extract the mac address in probe

Address_type addr=probe.addr2 ()；

Judge that the mac address captured occurred the most

AddrSet_type::iterator it=addr_set.find (addr)

If this mac address did not occurred, then mac address is stored in set

addr_set.insert(addr)；

(4), wireless signal process: according to signal attenuation formula Los=32.44+20lgd+20lgf,

In formula, Los is propagation loss, and unit is dB；D is distance, and unit is Km；F is operating frequency, and unit is MHz；

Signal intensity in detection region is modeled, it is thus achieved that wireless exploration equipment is to the sky of the spacing being calibrated equipment Between model, spatial modeling should model with wireless exploration equipment position, and image division angle is at [45 °, 180 °]；

Wherein spatial model to set up process as follows:

A, for plane, environment to be detected being evenly divided into m n with ground and take advantage of the net region of n, n typically takes 1m extremely 2m。

B, determine radio detection equipment position in a model, its storage format be (x, y).Wherein x, y are respectively grid At the coordinate figure of plane and as a reference point.

C, signal intensity step (3) gathered are by being stored as following data form after step (4) process: (int X, intY,struct MAC MAC)

(5), target is slightly mated: sets up the camera head space coordinates to the spacing of pedestrian, divides monitoring image and exist Corresponding domain in space, mates the detection target of camera head with the detection target of wireless exploration equipment, it is thus achieved that same Target sequence number corresponding relation in two data fields；

Wherein the determination process of camera head respective coordinates is as follows:

A, process distortion: distortion is that the shape of the optical imaging device such as lens having camera head causes, and has the most abnormal Become and radial distortion.

Distortion correction formula:

$\left[\begin{array}{c}{x}_p\\ y_p\end{array}\right]=(1+k_1{r}^2+k_2{r}^4+k_3{r}^6)\left[\begin{array}{c}{x}_d\\ y_d\end{array}\right]+\left[\begin{array}{c}2p_1{x}_d{y}_d+p_2(r^2+2{x_d}^2)\\ p_1(r^2+2{x_y}^2)+2p_2{x}_d{y}_d\end{array}\right]$

(x in formula_p, y_p) represent picture point normalization coordinate ideally, (x_d,y_d) indicate the picture point of distortion Normalization coordinate, the lens radius that r currently puts, k₁, k₂, k₃Determine the degree of radial distortion, and its value is Taylor's level respectively First 3 of number expansion.

B, three-dimensional scaling: the scene of video camera shooting is carried out space demarcation.

According to camera imaging model:

$x=\frac{{\mathrm{fX}}_c}{Z_c}$

$y=\frac{{\mathrm{fY}}_c}{Z_c}$

Wherein x, y are the coordinate at image midpoint, band X_c, Y_c, Z_cBeing respectively the coordinate of the point in space, f is focal length.

Corresponding relation between image coordinate system and physical coordinates mooring points is as follows:

$\left[\begin{array}{c}u\\ v\\ 1\end{array}\right]=\left[\begin{array}{ccc}\frac{1}{dx}& 0& u_0\\ 0& \frac{1}{dy}& v_0\\ 0& 0& 1\end{array}\right]\left[\begin{array}{c}x\\ y\\ 1\end{array}\right]$

Wherein u, v are the point of image coordinate system, and x, y are the point of physical coordinates system, u₀, v₀Initial point for image coordinate system.

The relation of camera coordinate system and world coordinate system is as follows:

$\left[\begin{array}{c}{X}_c\\ Y_c\\ Z_c\\ 1\end{array}\right]=\left[\begin{array}{cc}R& t\\ 0^T& 1\end{array}\right]\left[\begin{array}{c}{X}_w\\ Y_w\\ Y_w\\ 1\end{array}\right]$

Point under wherein subscript c represents camera coordinate system, subscript w represents the point under world coordinate system.R is camera Spin matrix, t is the translation vector of camera.

$\begin{array}{c}{Z}_c\left[\begin{array}{c}u\\ v\\ 1\end{array}\right]=\left[\begin{array}{ccc}\frac{1}{dx}& 0& u_0\\ 0& \frac{1}{dy}& v_0\\ 0& 0& 1\end{array}\right]\left[\begin{array}{cccc}f& 0& 0& 0\\ 0& f& 0& 0\\ 0& 0& 1& 0\end{array}\right]\left[\begin{array}{cc}R& t\\ 0^T& 1\end{array}\right]\left[\begin{array}{c}{X}_w\\ Y_w\\ Y_w\\ 1\end{array}\right]\\ =\left[\begin{array}{cccc}{f}_x& 0& u_0& 0\\ 0& f_y& v_0& 0\\ 0& 0& 1& 0\end{array}\right]\left[\begin{array}{cc}R& t\\ 0^T& 1\end{array}\right]\left[\begin{array}{c}{X}_w\\ Y_w\\ Y_w\\ 1\end{array}\right]=M_1{M}_2{M}_W={\mathrm{MX}}_W\end{array}$

Wherein M is the projection matrix of image.

C, in the picture delimitation ground grid

Draw grid in the picture so that it is corresponding relation is consistent with the coordinate system set in step (2), and representation is (x, the block position on ground during y) wherein x, y distinguish representation space model.Determine that two dimensional image is corresponding with the position of threedimensional model to close System.

D, image store data, is space coordinates (x, value y), and look into by the target designation detected in picture pick-up device Look for the target that the value of synchronization wireless exploration equipment and x, y is similar.

Meet:

|x_i-x_j| ＜ 2

|y_i-y_j| ＜ 2

Wherein x_i, y_iRepresent the target location that wireless exploration equipment obtains.I takes 1,2,3 ...

x_j, y_jRepresent the target location that wireless exploration equipment obtains.J takes 1,2,3 ...

Object matching matrix M is set to the corresponding relation characterizing between target, if meeting two above inequality, then in matrix Element M_{I, j}=1, it is otherwise 0.

(6), target essence coupling: according to the movement relation of front and back's frame target, detection target and the nothing to camera head simultaneously The detection target of line detecting devices carries out Kalman in the position of subsequent time and estimates, it is thus achieved that its moving displacement vector, according to position The detection target of confidence breath and motion vector assessment camera head and the matching degree of the detection target of wireless exploration equipment, arranged Whether two targets are the matching value of same target, and the allowable error of matching value not should be greater than 3m；

A, use Kalman filtering determine target travel, wherein state equation:

X (k+1)=A (k+1, k) X (k)+w (k)

Observational equation:

Z (k)=H (k) X (k)+v (k)

In formula, X (k) is state vector；Z (k) is state vector；(k+1 k) is state-transition matrix to A；H (k) is observation Matrix；W (k) is system noise vector；V (k) is observation noise vector.

W (k), V (k) usually assume that as 0 orthogonal average white Gaussian noise vectorial, and their covariance matrix is:

$E=\&lsqb;w\left(k\right)w^T\left(i\right)\&rsqb;=\left\{\begin{array}{c}Q\left(k\right),i=k\\ 0,i\&NotEqual;k\end{array}\right.$

$E=\&lsqb;v\left(k\right)v^T\left(i\right)\&rsqb;=\left\{\begin{array}{c}R\left(k\right),i=k\\ 0,i\&NotEqual;k\end{array}\right.$

To all of k and i,

E=[w (k) v^T(i)]=0

Predictive equation group is:

X ' (k+1 | k)=A (k+1 | k) X ' (k | k)

P (k+1 | k)=A (k+1 | k) P ' (k | k)

A^T(k+1, k)=Q (k)

Renewal equation group

K (k+1)=P (k+1 | k) H^T(k+1)[H(k+1)P(k+1|k)H^T(k+1)+R(k+1)]^-1X′(k+1|k+1)

=X ' (k+1 | k)+K (k+1) [Z (k+1)-H (k+1) X ' (k+1 | k)] ^ (-1)

P (k+1 | k+1)=[I-K (k+1) H (k+1)] P (k+1 | k)

The position that subsequent time target occurs can be constantly predicted by predictive equation.

B, set up vector according to current time position (x1, y1) and subsequent time position (x2, y2)

K1=[(x2-x1), (y2-y1)]

The vector device predicted with wireless exploration

K2=[(x ' 2-x ' 1), (y ' 2-y ' 1)]

Comparing, judgement schematics is:

$S=\sqrt{\&lsqb;{(x_2-x_1)}^2-{({x^{\&prime;}}_2-{x^{\&prime;}}_1)}^2\&rsqb;+\&lsqb;{(y_2-y_1)}^2-{({y^{\&prime;}}_2-{y^{\&prime;}}_1)}^2\&rsqb;}$

Setting threshold value m, < during m, target can merge S, and m then needs to be set according to scene.

(7), conclude the data detecting region internal object, remove repetition target, integrate detection target and the nothing of camera head No. ID of the detection target of line detecting devices, and unique ID number and real-time position information are stored in chained list.Target after merging Form:

(struct ID ID, int x, int y, struct MAC mac, point2f image)

Such as: A0007,21,15, F0:25:B7:4C:D5:D8,215.4,107.5

Although the present invention is open the most as above with preferred embodiment, but it is not limited to the present invention, any is familiar with this The people of technology, without departing from the spirit and scope of the present invention, can do various change and modification, the therefore protection of the present invention Scope should be with being as the criterion that claims are defined.

Claims (7)

1. view-based access control model and wireless aware combine pedestrian detection and statistical method, it is characterised in that comprise the following steps:

(1), monitoring image collection: utilize camera head to gather video monitoring image, the image obtained is carried out size change over；

(2), Image Information Processing: the imagery exploitation DPM method obtaining step (1) extracts the ROI in image, carries out color and leads to Road converts, and calculates feature pyramid, coupling master cast and submodel parts, calculating unit score, screens and demarcates score and surpass Cross the target area setting threshold value, it is thus achieved that the coordinate of the scope box of every frame line people's quantity and picture position, pedestrian place；

(3), wireless signal acquiring: what pedestrian was carried by the wireless exploration equipment that utilization is arranged in detection region has wireless communication The calibration facility of number transmission-receiving function makes requests on frame search, uses libtins to obtain claim frame data, the mac ground of analyzing device Location and signal intensity；

(4), wireless signal process: according to signal attenuation formula Los=32.44+20lgd+20lgf,

In formula, Los is propagation loss, and unit is dB；D is distance, and unit is Km；F is operating frequency, and unit is MHz；

Signal intensity in detection region is modeled, it is thus achieved that wireless exploration equipment is to the spatial mode of the spacing being calibrated equipment Type；

(5), target is slightly mated: sets up the camera head space coordinates to the spacing of pedestrian, divides monitoring image in space In corresponding domain, the detection target of camera head is mated with the detection target of wireless exploration equipment, it is thus achieved that same target Sequence number corresponding relation in two data fields；

(6), target essence coupling: according to the movement relation of front and back's frame target, detection target and the wireless spy to camera head simultaneously The detection target of measurement equipment carries out Kalman in the position of subsequent time and estimates, it is thus achieved that its moving displacement vector, believes according to position The matching degree of the detection target of breath and the detection target of motion vector assessment camera head and wireless exploration equipment, arranges two mesh Whether mark is the matching value of same target；

(7), conclude the data detecting region internal object, remove repetition target, integrate the detection target of camera head and wireless spy No. ID of the detection target of measurement equipment, and unique ID number and real-time position information are stored in chained list.

Pedestrian detection that view-based access control model the most according to claim 1 and wireless aware combine and statistical method, its feature Being, the transforming image dimension in described step (1) is suitable for VGA-QVGA.

Pedestrian detection that view-based access control model the most according to claim 1 and wireless aware combine and statistical method, its feature Being, the color channel in described step (2) is transformed to be converted to RGB triple channel information single-channel data, described step (2) In master cast quantity be 2, submodel number of components is 5, and the feature pyramid iteration number of plies is 10-48 layer, score threshold scope Setting value is [-1.8f, 1.0f].

Pedestrian detection that view-based access control model the most according to claim 1 and wireless aware combine and statistical method, its feature Being, the calibration facility in described step (3) is wireless network card or wireless router.

5. the pedestrian detection combined according to the view-based access control model described in claim 1 or 4 and wireless aware and statistical method, it is special Levying and be, the number of described calibration facility is more than 2.

Pedestrian detection that view-based access control model the most according to claim 1 and wireless aware combine and statistical method, its feature Being, in described step (4), spatial modeling should model with wireless exploration equipment position, image division angle [45 °, 180°]。

Pedestrian detection that view-based access control model the most according to claim 1 and wireless aware combine and statistical method, its feature Being, the allowable error of the matching value in described step (6) not should be greater than 3m.