Specific embodiment
The invention will be further described with the following Examples.
Referring to Fig. 1, Fig. 2, a kind of safe examination system on bus of the present embodiment, including hiding article detection apparatus
1st, microprocessor 2, flammable explosive gas monitor 3, alarm 4, the microprocessor 2 fill respectively with the concealment Articles detecting
Put 1, the flammable explosive gas monitor 3, the alarm 4 is connected;The concealment article detection apparatus 1 is used to detect public affairs
Hand over whether the vehicle personnel that get on the bus conceal dangerous article.
Preferably, the microprocessor 2 is set on the operation console of bus.
Preferably, the alarm 4 is set to bus railway carriage inner top, and reset key is arranged on the operation of bus
On platform.
The above embodiment of the present invention the passenger of upper bus can carry out the detection of carry-on articles, while monitor public affairs to every
The gas-condition in compartment is handed over, once detecting dangerous object, then microprocessor, which can send out control signal, makes alarm work;
Have many advantages, such as to monitor and save the time automatically.
The concealment article detection apparatus 1 includes:
(1) scan module 11 obtain original passive millimeter wave image for carrying out millimeter wave scanning to tested personnel;
(2) image processing module 12 obtain target image for carrying out processing to the original passive millimeter wave image;
(3) human region detection module 14 for being based on the target image, passes through the anticipation that whether there is to human body
It is disconnected, human region detection is carried out, obtains human region;
(4) concealment Articles detecting module 15, in the human region, using Canny boundary operators and confidence area
Between combine mixing segmentation method, to concealment article be detected, and the concealment article region detected is marked;
(5) item identification module is hidden, for concealment article to be identified.
This preferred embodiment devises the module architectures of concealment article detection apparatus 1, realizes concealment article detection apparatus 1
Detection to dangerous goods.
Preferably, described image processing module 12 includes:
(1) binarization unit obtains binary image, packet for carrying out binaryzation to the original passive millimeter wave image
It includes:The original passive millimeter wave image is subjected to smooth area P1, transition region P2With marginal zone P3Region division, use flat
Measurement of the square singular value as region energy, and definition region division formula is:
Wherein, v (i) is the observation at pixel i in original passive millimeter wave image, and Q (i) is corresponding original passive milli
Square singular value in metric wave image at pixel i,Mean value for square singular value;
For smooth area P1, given threshold T1, T1For smooth area P1The average value of all pixels gray value, by smooth area P1's
Each grey scale pixel value and threshold value T1It is compared, if more than T1, then value is 255, and otherwise value is 0;For transition region P2With
Marginal zone P3, transition region P2With marginal zone P3In (2a+1) x (2a+1) window centered on each pixel i, a ∈ [1,3], respectively
Pixel threshold value T2(i) be defined as maximum gradation value and minimum gradation value in the pixel window and half, by transition region P2And side
Edge area P3Each grey scale pixel value and respective threshold T2(i) compare, if more than T2(i), then value is 255, and otherwise value is 0;
(2) pretreatment unit, for carrying out surely picture and denoising to binary image to obtain preliminary denoising image,
Including:
A, image is surely as subelement, flating caused by for eliminating people during exercise, including:
(1) it when carrying out millimeter wave scanning to checked object, selectes first frame image and is used as with reference to frame, reference frame is divided into
Three regions 1 of non-overlapping copies, 2,3, K represent picture traverse, and G represents picture altitude, according to side clockwise since image upper left
To region 1,2,3 is followed successively by, the size in region 1,2 is 0.5K × 0.5G, and the size in region 3 is K × 0.5G;
(2) the image center location selection area A arrived in next frame scan0, A0Size be chosen to be 0.5K × 0.5G, press
According to step1 method by A0It is divided into three image subblock A1、A2、A3, A1And A2For estimating the local motion in vertical direction
Vector, A3For estimating the local motion vector in horizontal direction, A is enabled1、A2、A3It is searched most in 1,2,3 three region respectively
Good matching so as to estimate the global motion vector of image sequence, then carries out inverse motion compensation, eliminates image and obscures;
B, subregion denoising subelement, for smooth area P1, transition region P2With marginal zone P3Denoising is carried out respectively, if
Smooth area P1, transition region P2With marginal zone P3Denoising estimation be respectively G1(i)、G2(i) and G3(i), the denoising estimation in each region
Calculation formula be respectively:
In formula, N1For the total number of pixels of smooth area, V (j) is smooth area P1Gray value at point j, s1V (i) is passes through
The smooth area mean value at pixel i extracted after mean filter processing, s1V (j) is carries after being handled by mean filter
Smooth area mean value at the pixel j of taking-up;V (j) represents marginal zone P3In all pixel,For normaliztion constant, d is
Marginal zone P3The Gauss weighted euclidean distance of the annular feature vector of middle pixel i and pixel j, γ are the mark of gaussian kernel function
Accurate poor, μ is marginal zone P3The standard deviation of middle noise,And
C, denoising subelement again connects subregion denoising subelement, for using the full change based on division Bregman iteration
Point Denoising Algorithm is to by subregion denoising subelement, treated that passive millimeter wave image carries out further denoising, so as to further go
Except the white Gaussian noise in passive millimeter wave image.
This preferred embodiment setting image is surely as subelement, flating caused by can eliminating people during exercise;Two-value
Change in unit, the method combined using global threshold binaryzation and local threshold binarization is obtained under the premise of speed is ensured
Good effect;Passive millimeter wave image is made of the smooth image block of a small amount of approximate piecemeal, includes a large amount of redundancy letter
Breath according to this feature, sets pretreatment unit, passive millimeter wave image is divided into smooth area, mistake using singular value decomposition
Area and marginal zone are crossed, and subregion denoising subelement is set according to the respective feature in this three classes region, for smooth area, transition region
Denoising is carried out respectively with marginal zone, more compared with the concentration Image denoising algorithm of mainstream, can be obtained higher at present
BRISQUE improves the speed of calculating, and can significantly inhibit the influence of scan line noise and white Gaussian noise;Using again
Secondary denoising subelement with reference to subregion denoising subelement, further removes the white Gaussian noise in passive millimeter wave image, and raising is gone
It makes an uproar effect.
Preferably, the concealment Articles detecting module 15 selects suitable threshold value T when carrying out mixing segmentationyTo described
Target image carries out threshold process, and the confidence interval represents the tonal range of concealment article edge distribution, wherein including target
The global confidence interval at all edges of image is [int ((Vl+Vh)/2),Vh1, wherein int () be rounding operator, VlFor entire mesh
The lower bound of the minimum gradation value of the marginal point of logo image, VhThe upper bound of the maximum gradation value of marginal point for entire target image;
When calculating the confidence interval of each concealment article, using from Vl+ 1 starts point-by-point increased threshold value TgTarget image is divided
It cuts, until threshold value Tg=VhStop segmentation when -1, when occurring a new isolated area in prospect, defining corresponding position has
Another concealment article, the confidence interval for hiding article are defined as:Pz=[(Tg+Vh)/2,Vh], it is extracted using boundary operator hidden
Boundary curve in the confidence interval of article of hideing, it is P to define the region included in boundary curvec, the concealment object that finally obtains
The Accurate Segmentation region P of productyFor:
In formula,For by smaller threshold value Ty=(Tg+VhThe region that)/2 pair target image is split.
The method of mixing segmentation that this preferred embodiment is combined using Canny boundary operators and confidence interval is realized to hidden
It hides the segmentation of article, improves the speed and precision of segmentation.
Preferably, the human region detection module 14 can shield privacy places of human body, specifically include:
(1) gender of personnel is judged by face recognition, different privacy places is determined according to gender difference, drawn
Determine privacy area;
(2) Fuzzy Processing is carried out to determining privacy area, image obscures with the following method:
For privacy area any point (x, y), using functionDetermine the 5 × 5 of (x, y)
The value weight of neighborhood point carries out neighborhood weighted average to the point, the fuzzy value of the point is obtained, so as to reach blur effect.
This preferred embodiment to detect more hommization.
Finally it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention rather than the present invention is protected
The limitation of range is protected, although being explained in detail with reference to preferred embodiment to the present invention, those of ordinary skill in the art should
Work as understanding, technical scheme of the present invention can be modified or replaced equivalently, without departing from the reality of technical solution of the present invention
Matter and range.