CN100582650C - Gas leakage source searching method based on vision noticing mechanism - Google Patents

Abstract

The invention discloses a gas leakage source searching method based on a visual attention mechanism. In the gas leakage source searching process, the visual attention mechanism is inducted to lock a device which is easily leaked in a scene, thereby conducting a robot to fast and accurately prove the position of the gas leakage source. The realizing form is to adopt the forms of intensive collection on the information of the scene, intensive judging, overall planning and testing one by one. The steps include: training and obtaining the combined weight coefficient of a compared mapping graph which can convexly display a device that is easy to leak; adopting a full-scanning mode to intensively obtain the panorama information of the scene; confirming the quantities, positions and PRI of suspicious regions; testing whether the suspicious regions is the gas leakage source or not one by one according to a recursive optimization method. The invention adopts the modes of intensive collection, intensive judging and overall planning, is not affected by the situation of a flow field, can avoid a searching target to be disjointed with the actual demand, trains and obtains the compared mapping graph which can convexly display the device that is easy to leak and has the characteristics of high searching efficiency, high adaptation, high application value and low misjudging rate.

Description

Gas leakage source searching method based on vision noticing mechanism

Technical field

The present invention relates to a kind of method of toxic gas Leakage Detection.Particularly relate to a kind of searching efficiency height, applicability is strong, the high and low gas leakage source searching method based on vision noticing mechanism of False Rate of using value.

Background technology

Along with industrialized development, toxic gas has become the danger source that people have to face in producing and living.Oil and chemical industry, dangerous material accumulating, the attack of terrorism, garbage loading embeading etc. all are places that toxic gas takes place and leakage is comparatively general.Therefore, technical research and the application at search of toxic gas source of leaks and location just seems very necessary.On the one hand, its early warning and emergency processing of can be this type of accident provides technical support; On the other hand, for assuring the safety for life and property of the people, promote socioeconomic harmonious development to have great realistic meaning.

Traditional gas leakage source searching method is mainly finished in two steps.At first, should find gas leakage information as early as possible; Secondly, follow the trail of gas leakage information apace, promptly plume is found and the plume tracking.Aspect the plume discovery, generally adopt the random search mode at present, as method (periodicals: IEEE SensorsJournal such as " Z " (Zigzag) font motion, helical motions; Author: Hayes A T, Martinoli A, Goodman R M; Publish days: 2002; Title of article: Distributed odor source localization, the page number: 260-271).But, these easily method all be based on the identical hypothesis of probability that source of leaks appears in All Ranges in the scene, this hypothesis obviously is not inconsistent with actual conditions because the probability of source of leaks to occur big in some zone in the scene, some is then possible little.Therefore, there is certain blindness in above-mentioned random device.Aspect the plume tracking, at present typical tracing algorithm mainly contains two big classes: chemotaxis (Chemotaxis) method and wind taxis (Anemotaxis) method.Wherein, the chemotaxis method mainly is gas concentration (or concentration gradient) the tracking of information gas plume that obtains according to gas sensor, and finally finds odor source (periodical: Robotics and Autonomous Systems; Author: Russell R A, Alireza B H, ShepherdRL, et al; Publish days: 2003; Title of article: A Comparison of Reactive Robot ChemotaxisAlgorithms, the page number: 83-97).These class methods only can be successful in the plume environment of continuous distribution owing to the restriction that is subjected to sensor release time (how at tens seconds), only just may occur after indoor (mean wind speed is less than 20cm/s) and the long-time leakage of gas and such plume is general.But in actual plume,, thereby cause the concentration generation thermal agitation of plume inside, cause robot to be easy to be absorbed in local optimum and search failure because turbulent flow causes that the whirlpool of different scale can " tear the gas plume " and become a lot of filaments.Wind taxis method be when robot perceives gas plume information just along upwind follow the trail of and find odor source (periodical: IEEE International Conference onMechatronics and Automation; Author: Lu T F, Liang C; Publish days: 2006; Title of article: Novel wind sensor for robotic chemical plume tracking, the page number: 933-938).Mostly these class methods are to have carried out experimental verification under metastable artificial wind field (as wind-tunnel or fan) condition at present.And in actual wind field, the instability of wind direction occurs often, and this produces in various degree influence to the search effect of wind taxis method inevitably.In addition, above-mentioned method for searching based on non-visual sensing information generally can only utilize local message (as local concentration or wind speed/wind direction information) planning local path, a search task often needs repeatedly (timesharing) to gather, multistep planning just can be finished, therefore, the low shortcoming of these method ubiquity searching efficiencies.

More than analyze as can be seen, it is the subject matter that present traditional gas leakage source searching method faces that poor reliability, searching efficiency hang down.The survey showed that for a large amount of gas leakage accidents, the equipment that easily takes place to leak mainly is divided into 10 classes, i.e. pipeline, flexible connector, filtrator, valve, pressure vessel or reactor, pump, compressor, storage tank, pressurization or frozen gas container and exhaust stack (Beijing: Chemical Industry Press; The author: wangkai is complete, Shao Hui; Publish days: 2004; Works exercise question: accident theory and analytical technology).By discovering that these equipment all have the characteristics of himself in shape,, then can greatly improve the efficient and the success ratio of search if can in search process, give special concern to the equipment that these easily take place to leak by vision.For example, find the stage, can lock some easy equipment that leakage takes place in the scene in advance by vision, and at first the zone at these equipment places be searched that this can effectively overcome the blindness of random search to a certain extent at the gas plume; In the plume tracing process, because gas concentration field and wind field have complex changeable characteristics, cause chemotaxis method and all possible complete failure of wind taxis method in some cases, so, vision also can be next step judgement of searching direction useful clue is provided.Obviously, equipment is leaked in the easy generation that goes out by a large amount of gas leakage accidents investigation statisticses can provide important searching for clues, but at present both at home and abroad technical literature do not see the method for searching report that leaks equipment specially at concrete easy generation as yet.

Recent years, external minority scientific research personnel begins to attempt utilizing visual information to search gas leakage source, and obtains some preliminary achievements.People such as Kowadlo in 2006 propose to assist sense of smell to search (periodical: IEEE International Conference on Robotics and Automation with " crackle " as the visual signature of gas leakage source; Author: Kowadlo G, David R, Russell R A; Publish days: 2006; Title of article: Bi-modal Search usingComplementary Sensing (Olfaction/Vision) for Odour Source Localisation, the page number: 2041-2046).Obviously, this method only is applicable to that robot is nearer and leak the bigger situation in crack apart from source of leaks.In the same year, whether people such as Ishida also propose by the color discrimination object is suspicious object, but this method False Rate is higher, therefore is difficult to be applied to actual search task (periodical: Autonomous Robots; Author: Ishida H, Tanaka H, Taniguchi H; Publish days: 2006; Title of article: Mobile robot navigation using visionand olfaction to search for a gas/odor source, the page number: 231-238).In addition, often there are a plurality of suspicious object in next scene of normal conditions, plan that so how search path finishes the search task as early as possible, at present the also relevant achievement report of technical literature both at home and abroad.The existing method for searching that utilizes visual information occurs multiple goal inevitably and repeats the search problem owing to not planning search path, makes the efficient of these method for searching reduce greatly.

Analyze the technology that has document and report as can be seen, the application specific aim is strong, False Rate is high, searching efficiency is low is the present problem that mainly exists based on the method for searching of visual information.

Summary of the invention

Technical matters to be solved by this invention is, a kind of employing task-driven vision noticing mechanism is provided, judge the position of the suspicious object (leakage equipment easily takes place) in the scene rapidly, and the more excellent robot searching path of overall planning, finish the gas leakage source searching method based on vision noticing mechanism of the search task of gas leakage source quickly and accurately.

The technical solution adopted in the present invention is: a kind of gas leakage source searching method based on vision noticing mechanism, in the gas leakage source searching process, introduce vision noticing mechanism, the equipment of leakage easily takes place in the locking scene, it is suspicious target, thereby guided robot is verified the position of gas leakage source quickly and accurately, and specific implementation is the concentrated mode of gathering scene information, concentrating judgement, making overall planning, investigate one by one of employing, and concrete steps are as follows:

(1) training is obtained and can effectively be highlighted the contrast mapping graph merging weights coefficient that equipment is leaked in easy generation,

The contrast mapping graph merging weights coefficient that equipment is easily leaked in described acquisition is to adopt following formula:

$w(t+1)=w\left(t\right)+\mathrm{\η}\frac{{\stackrel{-}{M}}_{\mathrm{in}}-{\stackrel{-}{M}}_{\mathrm{out}}}{N}$

Wherein, η is a learning coefficient, M _InBe the average significantly value in source of leaks zone in the contrast mapping graph, M _OutBe the average significantly value that other zone source of leaks zone outside is removed in contrast in the mapping graph, N is that on average remarkable value significantly is worth M more than or equal to the source of leaks zone leveling in other zone except that the source of leaks zone _InRegional number, w (t) is that the contrast mapping graph merges the weights coefficients;

(2) adopt the full scan mode to concentrate the overall picture information of obtaining scene,

Described employing full scan mode is concentrated the overall picture information of obtaining scene, at first area size in this scene is surveyed by laser or sonac, the drive machines people moves to the geometric center of range coverage then, and this center is decided to be the start position of search, the mobile robot drives The Cloud Terrace and gathers the local scene image of a width of cloth every the field angle of setting, by the tested scene overall picture of several local scene image constructions;

(3) determine number, position and the priority of suspicious region,

The number of described definite suspicious region, position and priority comprise: at first, to local scene image sample, Filtering Processing, obtain the multiple dimensioned image information of many features; Secondly, adopt central authorities-peripheral difference operation to obtain many features and multiple dimensioned contrast mapping graph; Then, the contrast mapping graph is merged respectively with after the weights multiplication of distinct device, obtain locating the remarkable figure of relevant device, if significantly the average significantly value in zone surpasses setting threshold among the figure, then judge and have corresponding suspect device in this zone, this regional location of mark, and on average significantly being worth with suspicious region as this regional priority; At last, the analysis result of comprehensively all local scene images is determined number, position and the priority of suspicious region in the scene;

Whether (four) investigate these suspicious regions one by one by the recursion optimization method is gas leakage sources.

Whether described to investigate these suspicious regions one by one by the recursion optimization method are gas leakage sources, comprise: after the some suspected locations of eliminating are not source of leaks, be the searching target that starting point is planned next step with this position just, the priority of the path employing suspicious region of planning and the ratio of path are as optimizing index.

Gas leakage source searching method based on vision noticing mechanism of the present invention has following characteristics:

1. searching efficiency height, the method for searching employing that the present invention proposes is concentrated and is gathered, concentrates the strategy of judging, making overall planning, in addition, the judgement time of suspicious object has also been shortened in the introducing of vision noticing mechanism greatly, therefore, its searching efficiency is apparently higher than traditional method for searching based on non-visual sensing information (as chemotaxis method and wind taxis method).

2. applicability is strong, and what the present invention proposed is subjected to the influence of flow field situation hardly based on the method for searching of visual information, and its applicability obviously is better than traditional method for searching based on non-visual sensing information.

3. using value height, the source of leaks method for searching that the present invention proposes has taken into full account the application demand of gas leakage source searching task, the equipment that reality is easily leaked, given to pay close attention to, avoid the disconnection of searching target and actual demand, made method for searching have higher actual application value.

4. False Rate is low, the equipment that the vision positioning method that the present invention proposes easily leaks each class, all determine key character and best scale by a large amount of sample image training, thereby contrasted mapping graph accordingly and merged weights, guarantee that the equipment that easily takes place to leak can highlight effectively in the competition mechanism of attention model, therefore, significantly reduced the False Rate of searching target.

Description of drawings

Fig. 1 is the gas leakage source searching process flow diagram based on vision noticing mechanism of the present invention;

Fig. 2 is the one-piece construction synoptic diagram of gas leakage source searching device;

The search path planning chart of Fig. 3 recursion optimizing.

Embodiment

Below in conjunction with embodiment and accompanying drawing the gas leakage source searching method based on vision noticing mechanism of the present invention is made a detailed description.

Gas leakage source searching method based on vision noticing mechanism of the present invention is by being realized by mobile robot and sensor groups gas leakage source searching device dimerous as shown in Figure 2.

At different actual application environment, wheeled, crawler type or wheel-track combined that the mobile robot can adopt.Sensor groups comprises sonac 4 or laser range sensor 3 (executing gram LMS 200 as Germany), vision sensor 1 (as SonyEVI-D100), olfactory sensor 2 (as IBRID MX6) etc.Sonac 4 or laser range sensor 3 are used for the robot range coverage is surveyed; Vision sensor 1 be used to obtain scene image information (for the scene overall picture information that makes collection more comprehensively, the camera setting height(from bottom) is apart from ground 1.3-1.5m, and base assembling (pitching and level) two-degree-of-freedom cradle head); Olfactory sensor 2 can provide the gas concentration information of robot current location.

Gas leakage source searching method based on vision noticing mechanism of the present invention, be in the gas leakage source searching process, introduce vision noticing mechanism, the equipment of leakage easily takes place in the locking scene, it is suspicious target, thereby guided robot is verified the position of gas leakage source quickly and accurately, and specific implementation is the concentrated mode of gathering scene information, concentrating judgement, making overall planning, investigate one by one of employing.The specific implementation process comprises the steps: as shown in Figure 1

(1) training is obtained and can effectively be highlighted the contrast mapping graph merging weights coefficient that equipment is leaked in easy generation;

The present invention adopts the consolidation strategy of feature and yardstick competition mechanism with priority, to highlight the specific objective in the scene, i.e. and task-driven vision noticing mechanism computation model.Because the contrast mapping graph has comprised feature and yardstick information, the priority (weights size) that therefore only needs to obtain the contrast mapping graph gets final product, and the method for obtaining can adopt the weights alternative manner of supervision.The design of weights alternative manner should be satisfied following requirement: target and peripheral difference are big and the priority of the few contrast mapping graph in the zone of similar difference to occur higher, and less to the weights of the little contrast mapping graph of outstanding target contribution, in merging process, just can not weaken the significance degree of searching target like this.In addition, in actual applications, the less contrast mapping graph of those priority needn't be asked for, calculated amount can be significantly reduced like this.Formula (1) has provided the contrast mapping graph weights iterative formula that the present invention proposes.

$w(t+1)=w\left(t\right)+\mathrm{\η}\frac{{\stackrel{-}{M}}_{\mathrm{in}}-{\stackrel{-}{M}}_{\mathrm{out}}}{N}---\left(1\right)$

Wherein, η is a learning coefficient, M _InBe the average significantly value in source of leaks zone in the contrast mapping graph, M _OutBe the average significantly value that other zone source of leaks zone outside is removed in contrast in the mapping graph, N is that on average remarkable value significantly is worth M more than or equal to the source of leaks zone leveling in other zone except that the source of leaks zone _InRegional number.By the training of great amount of samples image, can obtain many features of distinct device and multiple dimensioned contrast mapping graph and merge the weights coefficient.

Before search, at first to obtain many features of easy generation leakage equipment and multiple dimensioned contrast mapping graph and merge weights.The weights iterative formula that the present invention adopts formula (1) to provide is trained the great amount of samples image, obtains the contrast mapping graph merging weights coefficient that 10 classes are easily leaked equipment respectively.But it should be noted that a bit 10 classes leakage equipment easily take place often appear at different occasions, therefore, should select a class or a few class that leakage equipment easily takes place and investigate according to the demand of the task of search in actual applications, to improve the specific aim and the accuracy of searching.

(2) adopt the full scan mode to concentrate the overall picture information of obtaining scene;

When the mobile robot enters a certain scene, at first can survey robot range coverage size in this scene by laser or sonac, the drive machines people moves to range coverage geometric center roughly then, and this center is decided to be the start position of search, the overall picture information of scene is convenient to obtain in this position, also helps the search path of making overall planning simultaneously.Scene overall picture information is obtained the full scan mode that adopts, and promptly the mobile robot drives The Cloud Terrace and gathers the local scene image of a width of cloth every fixing field angle, so the scene overall picture is just by several local scene image constructions.

Present embodiment is to drive The Cloud Terrace to gather the local scene image (field angle of the camera system that is adopted is about 50 °) of a width of cloth every 45 °, therefore the scene overall picture is made of 8 width of cloth topographies, as adopt wide-angle lens, can suitably correspondingly increase The Cloud Terrace scanning visual angle at interval.

(3) determine number, position and the priority of suspicious region;

At first, to local scene image sample, Filtering Processing, obtain the multiple dimensioned image information of many features; Secondly, adopt central authorities-peripheral difference operation to obtain many features and multiple dimensioned contrast mapping graph; Then, the contrast mapping graph is merged respectively with after the weights multiplication of distinct device, obtain searching the remarkable figure of relevant device, if significantly the average significantly value in zone surpasses setting threshold among the figure, then judge and have corresponding suspect device in this zone, this regional location of mark, and on average significantly being worth with suspicious region as this regional priority; At last, the analysis result of comprehensively all local scene images is determined number, position and the priority of suspicious region in the scene.

(1) for the multi-scale information in the abstract image, can adopt the gaussian pyramid form that original image is decomposed into 4 layers (corresponding yardsticks 0～4), wherein the 0th layer is original image, next tomographic image obtains by the last layer image being carried out low-pass filtering and interval sampling.

(2) pyramidal each layer is extracted a plurality of low-level visual features respectively, as brightness, color, towards etc.

Multiple dimensioned luminance picture I (σ _i)=(r (σ _i)+g (σ _i)+b (σ _i))/3

Wherein, σ _iFor yardstick (i=1,2 ... 4), r, g, b are three color component figure of true color image.

Multiple dimensioned color image is with two RG (σ _i), BY (σ _i) feature of flying up and down is to expression, the contrast of representative " red/green " and " indigo plant/Huang " color channel respectively, computing formula is as follows:

RG(σ _i)＝(r(σ _i)-g(σ _i))/max(r(σ _i)，g(σ _i)，b(σ _i)) (3)

BY(σ _i)＝(b(σ _i)-min(r(σ _i)，g(σ _i)))/max(r(σ _i)，g(σ _i)，b(σ _i)) (4)

At brightness pyramid image I (σ _i) on the basis, adopting direction is that the Gabor bank of filters of 0 °, 45 °, 90 ° and 135 ° is respectively to I (σ _i) carry out filtering, obtain multiple dimensioned towards image O (σ _i, θ), wherein θ ∈ [0 °, 45 °, 90 °, 135 °] is the Gabor filter direction.O (σ _i, computing formula θ) is as follows:

O(σ _i，θ)＝‖I(σ _i)*G ₀(θ)‖+‖I(σ _i)*G _π/2(θ)‖ (5)

Wherein,

Be the Gabor function, be expressed as:

f _x＝fcosθ，f _y＝fsinθ

In the formula, θ be towards, f represents change frequency, δ _xAnd δ _yThe reach of expression receptive field,

Corresponding multi-form receptive field:

The time be the symmetric form receptive field; The time be antisymmetry type receptive field.

(3) in order to obtain the comparative information on each feature dimensions, the present invention adopts " center-periphery is poor " extracting mechanism: at first defined feature pyramidal { 0,1, the 2} layer is central core c, { 3,4} is outer perisphere s, then central core and the periphery tomographic image corresponding pixel value of same category feature figure is subtracted each other, the center of finishing-periphery difference operation obtains contrasting mapping graph.The calculation combination of each category feature figure has 5 kinds: { (0-3), (1-3), (0-4), (1-4), (2-4) }.The little image of size at first will carry out interpolation processing when noting subtracting each other, and is the same with the size of two width of cloth images that guarantee to participate in computing.If note center-periphery difference is operating as Θ, then brightness, color and can try to achieve by formula (7)～(10) respectively towards the contrast mapping graph:

I(c，s)＝|I(c)ΘI(s)| (7)

RG(c，s)＝|RG(c)ΘRG(s)| (8)

BY(c，s)＝|BY(c)ΘBY(s)| (9)

O(c，s，θ)＝|O(c，θ)ΘO(s，θ)| (10)

Thus, totally 20 width of cloth are towards the contrast mapping graph can to obtain 5 width of cloth brightness contrast mapping graphs, 10 width of cloth color contrast mapping graphs and 4 directions, and totally 35 width of cloth contrast mapping graph.

(4) orientate example as with valve type equipment and describe suspicious region number, position and Determination of priority process in detail.The valve type equipment weights coefficient weighting that utilizes training to obtain merges 35 width of cloth contrast mapping graph, obtains searching the remarkable figure of this kind equipment.This remarkable figure is divided into different marking areas, calculate the average significantly value of zones of different, and compare with valve type judgment threshold given in advance (this threshold value is determined through a large amount of experiments), if average significantly value is greater than threshold value then think that this zone is a suspicious region, mark should the zone the position, and with the average significantly value of suspicious region as this regional priority.

Similar with above-mentioned localization method, try to achieve the suspicious region number, position and the priority that exist other classification easily to leak equipment in the Same Scene respectively.Merge these suspicious regions, obtain number, position and the priority in suspicious object zone total in the scene.

Whether (four) investigate these suspicious regions one by one by the recursion optimization method is gas leakage sources.

As only there being a suspicious region in the scene, directly the drive machines people searches to this zone.When occurring a plurality of may leak regional in the scene, just be necessary the search path of making overall planning, to improve searching efficiency.Can find source of leaks owing to can't determine which time investigation in advance, therefore, the present invention proposes to adopt the search strategy of recursion optimizing, promptly after the some suspected locations of eliminating are not source of leaks, be the searching target that starting point is planned next step just with this position, require the path of planning to satisfy certain optimization index, the priority (average significantly value) that the present invention adopts suspicious region and the ratio of path are as the optimization index.

Be the search process that example specifies the recursion optimizing there to be three suspicious target areas in the scene below, O is the scene geometric center among Fig. 3, and three target areas are labeled as A, B, C, supposes that their average significantly value is respectively P _A=2.7, P _B=1.8, P _C=1.5; AB line segment distance L _ABBe 10 meters, AC line segment distance L _ACIt is 6.5 meters.At first, drive machines people investigates to the highest suspicious region of priority (A position), if the output of olfactory sensor surpasses given threshold value, then thinks and finds gas leakage source, finishes the search task; If there is not gas leakage source in the A position, calculate other average significantly value of not investigating the zone and the ratio of path, i.e. P _B/ L _AB=0.18, P _C/ L _AC=0.23, because, P _B/ L _AB＞P _C/ L _AC, determine that therefore C is next step region of search, there is not gas leakage source as the C position yet, drive machines people investigates B.Should be noted in the discussion above that in the path planning process and should reject the zone of having investigated, avoid repeat search.

Claims (1)

1. gas leakage source searching method based on vision noticing mechanism, it is characterized in that, in the gas leakage source searching process, introduce vision noticing mechanism, the equipment of leakage easily takes place in the locking scene, promptly suspicious target, thus guided robot is verified the position of gas leakage source quickly and accurately, specific implementation is the concentrated mode of gathering scene information, concentrating judgement, making overall planning, investigate one by one of employing, and concrete steps are as follows:

(1) training is obtained and can effectively be highlighted the contrast mapping graph merging weights coefficient that equipment is leaked in easy generation,

The contrast mapping graph merging weights coefficient that equipment is easily leaked in described acquisition is to adopt following formula:

$w(t+1)=w\left(t\right)+\mathrm{\η}\frac{{\stackrel{\‾}{M}}_{\mathrm{in}}-{\stackrel{\‾}{M}}_{\mathrm{out}}}{N}$

Wherein, η is a learning coefficient, M _InBe the average significantly value in source of leaks zone in the contrast mapping graph, M _OutBe the average significantly value that other zone source of leaks zone outside is removed in contrast in the mapping graph, N is that on average remarkable value significantly is worth M more than or equal to the source of leaks zone leveling in other zone except that the source of leaks zone _InRegional number, w (t) is that the contrast mapping graph merges the weights coefficients;

(2) adopt the full scan mode to concentrate the overall picture information of obtaining scene,

Described employing full scan mode is concentrated the overall picture information of obtaining scene, at first area size in this scene is surveyed by laser or sonac, the drive machines people moves to the geometric center of range coverage then, and this center is decided to be the start position of search, the mobile robot drives The Cloud Terrace and gathers the local scene image of a width of cloth every the field angle of setting, by the tested scene overall picture of several local scene image constructions;

(3) determine number, position and the priority of suspicious region,

The number of described definite suspicious region, position and priority comprise: at first, to local scene image sample, Filtering Processing, obtain the multiple dimensioned image information of many features; Secondly, adopt central authorities-peripheral difference operation to obtain many features and multiple dimensioned contrast mapping graph; Then, the contrast mapping graph is merged respectively with after the weights multiplication of distinct device, obtain locating the remarkable figure of relevant device, if significantly the average significantly value in zone surpasses setting threshold among the figure, then judge and have corresponding suspect device in this zone, this regional location of mark, and on average significantly being worth with suspicious region as this regional priority; At last, the analysis result of comprehensively all local scene images is determined number, position and the priority of suspicious region in the scene;

Whether (four) investigate these suspicious regions one by one by the recursion optimization method is gas leakage sources.

Whether described to investigate these suspicious regions one by one by the recursion optimization method are gas leakage sources, comprise: after the some suspected locations of eliminating are not source of leaks, be the searching target that starting point is planned next step with this position just, the priority of the path employing suspicious region of planning and the ratio of path are as optimizing index.

Images