CN106125087B - Pedestrian tracting method in Dancing Robot room based on laser radar - Google Patents
Pedestrian tracting method in Dancing Robot room based on laser radar Download PDFInfo
- Publication number
- CN106125087B CN106125087B CN201610439857.2A CN201610439857A CN106125087B CN 106125087 B CN106125087 B CN 106125087B CN 201610439857 A CN201610439857 A CN 201610439857A CN 106125087 B CN106125087 B CN 106125087B
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- point
- cluster
- data
- laser
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
- G01S17/66—Tracking systems using electromagnetic waves other than radio waves
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
Abstract
The present invention proposes pedestrian tracting method in a kind of Dancing Robot room based on laser radar, includes the following steps：(1) laser data pre-processes；(2) the laser data cluster based on density；(3) pedestrian's identification based on laser collection point；(4) graphic software platform；(5) pedestrian tracking based on particle filter；(6) pedestrian movement track is drawn.The present invention realizes that the pedestrian of different distance different shape identifies using single laser radar, and can be to the pedestrian of the different motions pattern such as static, walking into line trace, and following range is wide, and real-time is good, and the accuracy of recognition and tracking is high.
Description
Technical field
The invention belongs to robotic technology fields, and in particular to the interior based on laser radar for Dancing Robot
Pedestrian tracting method.
Background technology
Robot technology achieves surprising development, the various occasions of human lives, such as medical treatment, family, meal between decades
Shop, hotel etc., with the presence of the robot for substituting people or making with indirect labor.Due to needing to coexist in many cases with people, because
This robot wants the safety of guarantor, working level more demanding while completion itself work.Active negotiation robot
Robot and the hight coordinate work of people can be realized under multisensor, improve the possibility coexisted with people.Typical application
Such as：In order to realize that Dancing Robot is danced in the case where being contacted with people with people, robot needs to observe the environment of surrounding, and
And the motor coordination of itself is carried out according to the movement locus of pedestrian, complete dance movement under the premise of ensureing other people safety.Cause
This, the motor coordination that the pedestrian in ambient enviroment is identified during dancing to robot with tracking has with trajectory planning
Significance.
The pedestrian tracting method of robot has very much, and the pedestrian tracking based on video is current most common method, but by
In the row for being influenced so that visual information is utilized to complete Dancing Robot by camera fields of view range, light, pedestrian's attitudes vibration
People's tracking becomes abnormal difficult, and especially when people is closer apart from video camera, video camera can only collect pedestrian topography.Therefore
The pedestrian tracking of view-based access control model is chiefly used in outdoor or indoor long distance pedestrian tracking.Realize that pedestrian tracking is using laser radar
Another common methods.Laser radar is a kind of new type measuring instrument table measured using laser technology, high with precision,
The features such as speed is fast, wide range of measurement.But the pedestrian tracting method typically cost based on laser radar is relatively high, and be easy by
Influence to algorithm makes real-time and accuracy rate not high.Patent CN102253391A discloses a kind of based on multilasered optical radar
Indoor pedestrian position and the movement locus of real-time tracking pedestrian can be effectively detected out in pedestrian target tracking.But the party
The radar of method is in a fixed position, and needs to demarcate multiple radars, and multiple radar costs are high, are not suitable for mobile dance
Step the pedestrian tracking of robot.
In conclusion being applied to main problem existing for the indoor pedestrian tracting method of Dancing Robot at present：It is based on
The pedestrian tracting method of video is difficult to meet short-range pedestrian tracking, and the pedestrian tracting method based on laser radar need to consider into
This, and meet real-time and accuracy requirement, therefore a low cost is researched and developed, real-time and practicability are good, and tracking accuracy rate is high
Indoor pedestrian tracting method have very high application value.
Invention content
It is an object of the invention to overcome the shortcomings and deficiencies of existing indoor pedestrian tracting method, provide a kind of based on sharp
Pedestrian tracting method in the Dancing Robot room of optical radar realizes the pedestrian of different distance different shape using single laser radar
Identification, and can be to the pedestrian of the different motions pattern such as static, walking into line trace, following range is wide, and real-time is good, identification with
The accuracy of track is high.
In order to achieve the above objectives, the technical proposal of the invention is realized in this way, a kind of dancing machine based on laser radar
Pedestrian tracting method in device people room, includes the following steps：
(1) laser data pre-processes：It is connected to the laser radar of robot host, collects the object scanned at certain angle
Range information of the body to radar；Host obtains laser initial data, is pre-processed to initial data, during preprocess method includes
Value filtering and coordinate system conversion；
(2) the laser data cluster based on density：The data obtained for step (1) are realized by density clustering
Environment is divided, and different objects are distinguished；
(3) pedestrian's identification based on laser collection point：To the click-through every trade people identification in the class cluster that clusters out, remove non-row
People's cluster, using pedestrian's cluster as the input data of pedestrian tracking；
(4) graphic software platform：The scan data that laser continuously acquires is patterned display, i.e., is converted laser data
At video flowing；One mapping is done to each collection point, and collection point is subjected to image magnification, is intuitively carried out convenient for more efficient
Tracking；
(5) pedestrian tracking based on particle filter, including：
(501) initial phase：
After completing pedestrian's identification, using the position of pedestrian's cluster as the initial position of pedestrian tracking, first initialization tracking
Device extracts the feature in rectangle frame, and provincial characteristics is described by the distribution of color probability density of target area, and to particle collection into
Row initialization；
(502) propagation stage：According to equation of transfer st=Ast-1+wt-1, calculate new particle collection and predict the position of new particle
It sets, wherein A is state-transition matrix, wt-1For Gaussian noise；
(503) decision phase：The measurement standard of the characteristic similarity of particle is chosen, the similarity calculated or score are i.e.
For the weight of corresponding particle, the weight of each particle is calculated, and is normalized；
(504) the state estimation stage：According to particle weights, the position of particle is summed according to weight to obtain the position of target
Output；
(505) the resampling stage：Resampling is carried out to particle according to weights of importance, obtains new particle collection；
(506) step (502) is repeated to step (505), realizes the tracking of moving object；
(6) pedestrian movement track is drawn：Pedestrian position is obtained to each frame imagery exploitation particle filter, calculates target's center
Position, and line is done at the center and former frame target's center, obtain pedestrian movement's trajectory diagram.
Further, the method for step (1) described medium filtering is：The collection point of laser radar belong to two-dimemsional number it is believed that
Number, collected data are a distance of the point to laser, to the range data of the m data point in pending data vertex neighborhood into
(1+m)/2 distance after sequence, is replaced the distance of current data point by row sequence.
Further, the method for step (1) the coordinate system conversion is：Coordinate system is converted, and initial data is filtered through intermediate value
Part measurement error is eliminated after wave, since initial data is to be converted for the ease of data analysis under polar coordinate system
To rectangular coordinate system, raw data format is：
sk=dk, k=0,1 ..., N (1),
Wherein, dkFor the distance of k-th of collection point；
It is converted the binary vector group into rectangular coordinate system：
uk=(xk,yk)T, k=0,1 ..., Ne(2),
Wherein, ukFor the corresponding rectangular co-ordinate in k-th of laser collection point, xkWith ykRespectively X-axis and Y axis coordinate, NeTo have
Imitate collection point number；；
If point P is k-th of collection point, the azimuth that laser radar obtains is θk, distance is ρk, angular resolution β, then θk
=k × β, acquiring the corresponding rectangular co-ordinates of P is：
Px=ρk cos(θk- 45 °) (3),
Py=ρk sin(θk- 45 °) (4),
Wherein, PxWith PyThe respectively X-axis and Y axis coordinate of point P, (θk- 45 °) be point P and X-axis angle.
Further, it in step (2), is clustered using DBSCAN algorithms, key step includes：
(201) all the points are labeled as unvisited；
(202) the random point p for taking unvisited, and it is labeled as visited；
(203) judge whether point p is core point, if the number to point of the point p distances less than ε is more than MinPts, p is core
Heart point continues to walk (4), and otherwise p is noise, returns to step (2)；
(204) p is core point, cluster C is built, and p is added to C, by PNIt is defined as the set put in ε-neighborhood of point p, after
Continuous step (5)；
(205) for PNMidpoint q repeats to walk (6) to PNFor sky；
(206) if the label of q is to be marked as visited first, then judge whether q is core
Point in ε-neighborhood of q is added to P by point if q is core pointNIf q is not the member of any cluster, q is added to C；
(207) output cluster C；
(208) step (202) to (207) to all the points are repeated and are labeled as visited；
Further, when above-mentioned DBSCAN algorithms carry out core point calculating, for k-th of collection point Pk, need Pk
With remaining all the points into the calculating of row distance, time complexity is high, therefore when carrying out core point calculating, only considers PkAround
Point and PkDistance, that is, calculate P(k-MinPts)To P(k+MinPts)Point in range is to PkDistance.
Further, the specific method is as follows for step (3) pedestrian's identification：
(301) point in the class cluster that clusters out is scanned first, obtains the information of each class cluster, including adopted in class cluster
Collect the number N of pointc, average distance D of the class cluster apart from laser radaravgAnd the position coordinates at the edge up and down of class cluster；
(302) coarse sizing is carried out according to the size of class cluster；
(303) fine screening is carried out to remaining cluster, is obtained laser radar first to the distance of the cluster, is denoted as r, enables r
=DavgIf pedestrian's waist width is Wped, WpedValue range be Wmin<Wped<Wmax；If using r as radius, by WpedAs arc
It is long, then there is following formula：
The corresponding central angle θ of sector region is calculated, formula is as follows：
The number P of collection point is calculated using formula (7) according to the corresponding central angle θ of the angular resolution β and cluster of laser radarc,
Calculation formula is as follows：
If the number of the corresponding point of pedestrian's cluster is denoted as Pnum, then PnumRange estimated value computational methods it is as follows：
Cluster after each cluster is judged according to the above method, removes non-pedestrian cluster, using pedestrian's cluster as row
The input data of people's tracking.
Compared with the existing technology, pedestrian tracking side in a kind of Dancing Robot room based on laser radar of the present invention
Method mainly has following advantage：
(1) the method for the present invention is utilized single laser radar and realizes indoor pedestrian tracking, compared to the pedestrian based on video
Tracking expands following range, and the pedestrian tracting method compared to multilasered optical radar reduces cost；
(2) present invention clusters original laser data, in order to promote the speed of cluster, according to laser data collection
Characteristic distributions are optimized original DBSCAN algorithms, are greatly improved cluster speed, improve efficiency of algorithm；
(3) present invention analyzes laser radar data collection point quantity and the relationship of object distance, it is proposed that base
In the distance of class cluster to laser radar and the pedestrian recognition method of pedestrian body width, this method recognition speed is fast, and accuracy rate is high；
(4) present invention in order to accurate and visual realizes pedestrian tracking, laser data progress image conversion is shown, by dispersion number
Strong point is converted to video data and is handled, and real-time is good, convenient for more intuitive efficient realization pedestrian tracking.
Description of the drawings
Fig. 1 is that the present invention is based on pedestrian tracting method flow charts in the Dancing Robot room of laser radar；
Fig. 2 is coordinate system transition diagram in the method for the present invention；
Fig. 3 is status diagram of the pedestrian of different location different shape in the method for the present invention under laser radar；
Fig. 4 is that the number of collection point in pedestrian's cluster in the method for the present invention estimates schematic diagram；
Fig. 5 (a) is the track path with Actual path comparison diagram of stationary state pedestrian in the method for the present invention；
Fig. 5 (b) is the track path and Actual path comparison diagram of the method for the present invention linear movement pedestrian；
Fig. 5 (c) is the track path with Actual path comparison diagram of curvilinear motion pedestrian in the method for the present invention.
Specific implementation mode
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the present invention can phase
Mutually combination.
Below in conjunction with embodiment and attached drawing, the present invention will be described in detail.
The present invention installs laser radar in Dancing Robot waist, obtains the range data scanned, and carry out to data
Pretreatment and clustering by laser data graphic software platform, and utilize particle by realizing pedestrian's identification to the analysis of class cluster
Filtering algorithm realizes pedestrian tracking；Fig. 1 is that the present invention is based on pedestrian tracting method flows in the Dancing Robot room of laser radar
Figure, the described method comprises the following steps：
The first step, laser data pretreatment：
Laser radar is positioned over to the waist location of Dancing Robot, apart from ground 1m, selects Hokuyo companies of Japan
URG laser radars, and host is connected to by USB interface, it is communicated using SCIP2.0 data protocols, collected is certain
Range information of the object scanned at a angle to radar；Host obtains laser original number by sending protocol instructions first
According to, in order to remove the measurement error of initial data and accelerate later stage arithmetic speed, initial data is pre-processed, pretreatment packet
Include medium filtering and coordinate system conversion；
(1) medium filtering：
The collection point of laser radar belongs to 2-D data signal, and collected data are distance of the point to laser, treat place
The range data of m data point in the data vertex neighborhood of reason is ranked up, and (1+m)/2 distance after sequence is replaced working as
The distance of preceding data point, wherein m take 5, just complete the Data correction of the point in this way；
(2) coordinate system is converted：
Initial data is through eliminating part measurement error after medium filtering, since initial data is under polar coordinate system,
It is converted to rectangular coordinate system for the ease of data analysis, raw data format is：
sk=dk, k=0,1 ..., N (1),
Wherein, dkIt is the coordinate system transition diagram in the method for the present invention for the distance of k-th of collection point, such as Fig. 2, A is
Sweep starting point, C are sweep stopping point, and B is the first effective collection point got；
It is converted the binary vector group into rectangular coordinate system：
uk=(xk,yk)T, k=0,1 ..., Ne(2),
Wherein, ukFor the corresponding rectangular co-ordinate in k-th of laser collection point, xkWith ykRespectively X-axis and Y axis coordinate, NeTo have
Imitate collection point number；
If the midpoints Fig. 2 P is k-th of collection point, the azimuth that laser radar obtains is θk, distance is ρk, angular resolution β,
Then θk=k × β, acquiring the corresponding rectangular co-ordinates of P is:
Px=ρk cos(θk- 45 °) (3),
Py=ρk sin(θk- 45 °) (4),
Wherein, PxWith PyThe respectively X-axis and Y axis coordinate of point P, (θk- 45 °) be point P and X-axis angle；
Second step：Laser data cluster based on density：
The data obtained for the first step realize environment segmentation by density clustering, and different objects are carried out area
Point；DBSCAN is a kind of density-based algorithms, for finding the dense Region in data set, and be not easy by noise and
The influence of outlier；The input parameter of DBSCAN algorithms is the density threshold MinPts of radius of neighbourhood ε and dense Region；If one
MinPts point is included at least in ε-neighborhood of a point, then the point is known as core point, the point composition one in core point and ε-neighborhood
A cluster merges the cluster belonging to these core points if containing multiple cores point in a cluster, obtains finally clustering knot
Fruit；
The key step that DBSCAN algorithms are clustered：
(1) all the points are labeled as unvisited；
(2) the random point p for taking unvisited, and it is labeled as visited；
(3) judge whether point p is core point, if the number to point of the point p distances less than ε is more than MinPts, p is core
Point continues to walk (4), and otherwise p is noise, returns to step (2)；
(4) p is core point, cluster C is built, and p is added to C, by PNIt is defined as the set put in ε-neighborhood of point p, is continued
It walks (5)；
(5) for PNMidpoint q repeats to walk (6) to PNFor sky；
(6) if the label of q is to be marked as visited first, then judge whether q is core point,
If q is core point, the point in ε-neighborhood of q is added to PNIf q is not the member of any cluster, q is added to C；
(7) output cluster C；
(8) step (2) to (7) to all the points are repeated and are labeled as visited；
Since the distribution of the increase of the distance arrived with laser radar scanning, point increasingly disperses, in order to more accurately
It is clustered, the present invention sets ε to 100mm, and MinPts is set as 5；According to original DBSCAN clustering algorithms, clustered
Journey time complexity is higher, is not achieved requirement of real-time, thus according to the data characteristics of laser collection point to DBSCAN algorithms into
Row optimization；
When original DBSCAN algorithms carry out core point calculating, for k-th of collection point Pk, need PkWith remaining institute
A little into the calculating of row distance, time complexity is high, since laser collection point is the point on sectoring face, the remoter point of distance
The possibility for belonging to same object is smaller, therefore when carrying out core point calculating, only considers PkAround put and PkDistance, i.e.,
Calculate P(k-5)To P(k+5)Point in range is to PkDistance, greatly reduced while not influencing Clustering Effect calculate the time,
Improve the speed of cluster；
Third walks：Pedestrian's identification based on laser collection point：
The laser beam of laser radar transmitting is covering of the fan, and pedestrian's cluster is generally arc or near linear, apart from laser radar
Distance directly affect scanning to object point quantity；Direction, angle, size all same but apart from two different objects,
More compared with the quantity of collection point on close object to laser radar, density is high, to the number of collection point on the object of laser radar farther out
Amount is few, and density is low；
Except pedestrian to laser radar distance influence collection point quantity in addition to, pedestrian during the motion with laser radar
Relative position can change, and the different shape of pedestrian can also have an impact the quantity of collection point；As Fig. 3 be different location not
With state of the pedestrian under laser radar of form, round rectangle represents the waist profile of pedestrian, five kinds of differences is shown in figure
The acquisition with the pedestrian of different shape different location is arrived in state of the pedestrian under laser radar under different shape, scanning
Point is also different, but can be to being adopted in the pedestrian body under different location different shape according to the width information of the waist location of pedestrian
The quantity of collection point is estimated, to carry out pedestrian's identification in multiple clusters；
The specific method is as follows for pedestrian's identification：
(1) point in the class cluster that clusters out is scanned first, obtains the information of each class cluster, including acquired in class cluster
The number N of pointc, average distance D of the class cluster apart from laser radaravgAnd the position coordinates at the edge up and down of class cluster；
(2) coarse sizing is carried out according to the size of class cluster, the cluster by the number at class cluster midpoint more than 200 is considered as non-pedestrian cluster；
(3) fine screening is carried out to remaining cluster, using pedestrian at a distance from laser radar and the body width of pedestrian
The number for the laser collection point reflected in pedestrian body is estimated；Fig. 4 is the number estimation signal of collection point in pedestrian's cluster
Figure obtains laser radar to the distance of the cluster, is denoted as r, enables r=D firstavgIf pedestrian's waist width is Wped, WpedValue
Ranging from Wmin<Wped<Wmax；If using r as radius, by WpedAs arc length, then there is following formula：
The corresponding central angle θ of sector region is calculated, formula is as follows：
The number P of collection point is calculated using formula (7) according to the corresponding central angle θ of the angular resolution β and cluster of laser radarc,
Then the number of the corresponding point of cluster is PcOr Pc+ 1, PcCalculation formula it is as follows：
If the number of the corresponding point of pedestrian's cluster is denoted as Pnum, then PnumRange estimated value computational methods it is as follows：
W is taken in the present inventionpedRanging from 200mm<Wped<500mm, to the cluster after each cluster according to the above method
Judged, remove non-pedestrian cluster, using pedestrian's cluster as the input data of pedestrian tracking；
4th step：Graphic software platform：
Since the laser collection point of acquisition relatively disperses, sweep spacing 100ms, the time is longer, therefore adjacent twice sweep
Obtained data variation is larger, in pedestrian body the position of collection point and deformation also become larger therewith；In consideration of it, laser is continuous
The scan data of acquisition is patterned display, i.e., laser data is converted into the video flowing that frequency acquisition is 10 frames/s；
The rectangular co-ordinate of each collection point is obtained after coordinate system is converted to the data scanned every time, in order to accelerate tracking
Coordinate data in rectangular coordinate system is reduced 20 times of displays, using image center location as the position of laser radar, i.e., by speed
Coordinate origin；One mapping is done to each collection point, and collection point has been subjected to image magnification, convenient for it is more efficient intuitively into
Line trace；
5th step：Pedestrian tracking based on particle filter：
(1) initial phase：
After completing pedestrian's identification, using the position of pedestrian's cluster as the initial position of pedestrian tracking, first initialization tracking
Device extracts the feature in rectangle frame, and is initialized to particle collection, particle number 100；
Provincial characteristics is described by the distribution of color probability density of target area, and method is as follows：
Discrete statistics is carried out to region to be tracked first and obtains the color histogram of RGB；For enhancing distribution of color can
By property, the pixel of distance center farther out is assigned to smaller weights, weight function is as follows：
Wherein, r is the distance for a little arriving target's center, if target area centre coordinate is Xc=(xc,yc), HxAnd HyFor target
The width and height in region, then target area size useThe coordinate of description, target area ith pixel is Xi=
(xi,yi), i=1, n, then current state XtColor distribution model be：
Wherein, n be current region number of pixels, m be histogram feature number, wherein m=256, f be normalization because
Son, u are that the color grade of histogram indexes, δ [h (Xi)-u] indicate X in target areaiWhether the color at position belongs to histogram
U-th of feature in figure, the value is 1 if belonging to, and is otherwise 0；
(2) propagation stage：According to equation of transfer st=Ast-1+wt-1, calculate new particle collection and predict the position of new particle
It sets, wherein A is state-transition matrix, wt-1For Gaussian noise；
(3) decision phase：The measurement standard of the characteristic similarity of particle is chosen, the similarity or score calculated are
The weight of corresponding particle, method are as follows：
Each particle calculates Color histogram distribution according to above-mentioned steps (1), and by the color histogram of itself and target area
Figure distribution is compared, and is weighed using Bhattacharyya distances, computational methods are as follows：
Wherein, p (Xt) it is Color histogram distribution at t-th particle position, pdFor the color histogram point of target area
Cloth；
The weighted value of each particle is obtained according to Bhattacharyya distances：
Wherein, σ is the Gauss difference of two squares, and value 0.1, the corresponding weights of particle are bigger, then the particle is got over target similitude
Greatly；
The weight of each particle is calculated, and is normalized；
(4) the state estimation stage：According to particle weights, the position for summing to obtain target according to weight to the position of particle is defeated
Go out；
(5) the resampling stage：Resampling is carried out to particle according to weights of importance, obtains new particle collection；
(6) step (2) to (5) is repeated, realizes the tracking of moving object；
6th step：It draws pedestrian movement track：
Pedestrian position is obtained to each frame imagery exploitation particle filter, calculates target's center position, and by the center with before
Line does in one frame target's center, obtains pedestrian movement's trajectory diagram；
In order to verify the present invention pedestrian tracking effect, to different distance, different shape, Different Exercise Mode pedestrian into
Row identification and tracking；If Fig. 5 (a) is the track path and Actual path comparison diagram of stationary state pedestrian in the method for the present invention,
Middle tracing positional fluctuates near pedestrian, does not occur relatively large deviation, and tracking effect is good；If Fig. 5 (b) is straight in the method for the present invention
Line moves the track path and Actual path comparison diagram of pedestrian, the posture in pedestrian movement and the relative position with laser radar
Constantly variation, experiment show that tracking result is accurate, and smaller error occurs in part path, but does not cause big shadow to tracking
It rings；If Fig. 5 (c) is the track path and Actual path comparison diagram of curvilinear motion pedestrian in the method for the present invention, due to the fortune of pedestrian
Dynamic direction changes very fast, and also relatively acutely, therefore it can be seen from the figure that fluctuation is more apparent for metamorphosis, but algorithm correct in time with
Track route；By three groups of experiments it is found that the track algorithm of the present embodiment to the pedestrian of movement can rapidly and accurately into line trace, and
Tracing deviation can be corrected in time.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
With within principle, any modification, equivalent replacement, improvement and so on should all be included in the protection scope of the present invention god.
Claims (6)
1. pedestrian tracting method in a kind of Dancing Robot room based on laser radar, which is characterized in that include the following steps：
(1) laser data pre-processes：It is connected to the laser radar of robot host, the object scanned at certain angle is collected and arrives
The range information of radar；Host obtains laser initial data, is pre-processed to initial data, and preprocess method includes intermediate value filter
Wave and coordinate system conversion；
(2) the laser data cluster based on density：The data obtained for step (1) realize environment by density clustering
Segmentation, different objects are distinguished；
(3) pedestrian's identification based on laser collection point：To the click-through every trade people identification in the class cluster that clusters out, remove non-pedestrian
Cluster, using pedestrian's cluster as the input data of pedestrian tracking；
(4) graphic software platform：The scan data that laser continuously acquires is patterned display, i.e., is converted into regarding by laser data
Frequency flows；One mapping is done to each collection point, and collection point is subjected to image magnification, convenient for it is more efficient intuitively carry out with
Track；
(5) pedestrian tracking based on particle filter, including：
(501) initial phase：
After completing pedestrian's identification, using the position of pedestrian's cluster as the initial position of pedestrian tracking, tracker is initialized first, is carried
The feature in rectangle frame is taken, provincial characteristics is described by the distribution of color probability density of target area, and is carried out just to particle collection
Beginningization；
(502) propagation stage：According to equation of transfer st=Ast-1+wt-1, it calculates new particle collection and predicts the position of new particle,
Middle A is state-transition matrix, wt-1For Gaussian noise；
(503) decision phase：The measurement standard of the characteristic similarity of particle is chosen, the similarity or score calculated are phase
The weight for answering particle, calculates the weight of each particle, and is normalized；
(504) the state estimation stage：According to particle weights, the position for summing to obtain target according to weight to the position of particle is defeated
Go out；
(505) the resampling stage：Resampling is carried out to particle according to weights of importance, obtains new particle collection；
(506) step (502) is repeated to step (505), realizes the tracking of moving object；
(6) pedestrian movement track is drawn：Pedestrian position is obtained to each frame imagery exploitation particle filter, calculates target's center position
It sets, and line is done at the center and former frame target's center, obtain pedestrian movement's trajectory diagram.
2. pedestrian tracting method in a kind of Dancing Robot room based on laser radar according to claim 1, feature
It is, the method for step (1) described medium filtering is：The collection point of laser radar belongs to 2-D data signal, collected number
According to being distance of the point to laser, the range data of the m data point in pending data vertex neighborhood is ranked up, will be sorted
(1+m)/2 distance afterwards replaces the distance of current data point.
3. pedestrian tracting method in a kind of Dancing Robot room based on laser radar according to claim 1 or 2, special
Sign is that the method for step (1) the coordinate system conversion is：Coordinate system is converted, and initial data after medium filtering through eliminating
Part measurement error, since initial data is to be converted to rectangular coordinate system for the ease of data analysis under polar coordinate system,
Raw data format is：
sk=dk, k=0,1 ..., N (1),
Wherein, dkFor the distance of k-th of collection point；
It is converted the binary vector group into rectangular coordinate system：
uk=(xk,yk)T, k=0,1 ..., Ne(2),
Wherein, ukFor the corresponding rectangular co-ordinate in k-th of laser collection point, xkWith ykRespectively X-axis and Y axis coordinate, NeEffectively to adopt
Collection point number；
If point P is k-th of collection point, the azimuth that laser radar obtains is θk, distance is ρk, angular resolution β, then θk=k ×
β, acquiring the corresponding rectangular co-ordinates of P is：
Px=ρkcos(θk- 45 °) (3),
Py=ρksin(θk- 45 °) (4),
Wherein, PxWith PyThe respectively X-axis and Y axis coordinate of point P, (θk- 45 °) be point P and X-axis angle.
4. pedestrian tracting method in a kind of Dancing Robot room based on laser radar according to claim 1, feature
It is, in step (2), is clustered using DBSCAN algorithms, input parameter is the density threshold of radius of neighbourhood ε and dense Region
MinPts, key step include：
(201) all the points are labeled as unvisited；
(202) the random point p for taking unvisited, and it is labeled as visited；
(203) judge whether point p is core point, if the number to point of the point p distances less than ε is more than MinPts, p is core
Point continues step (204), and otherwise p is noise, return to step (202)；
(204) p is core point, cluster C is built, and p is added to C, by PNIt is defined as the set put in ε-neighborhood of point p, continues step
(205)；
(205) for PNMidpoint q repeats step (206) to PNFor sky；
(206) if the label of q is to be marked as visited first, then judge whether q is core point, if q
For core point, the point in ε-neighborhood of q is added to PNIf q is not the member of any cluster, q is added to C；
(207) output cluster C；
(208) step (202) to (207) to all the points are repeated and are labeled as visited.
5. pedestrian tracting method in a kind of Dancing Robot room based on laser radar according to claim 4, feature
It is, when above-mentioned DBSCAN algorithms carry out core point calculating, for k-th of collection point Pk, need PkWith remaining all click-through
The calculating of row distance, time complexity is high, therefore when carrying out core point calculating, only considers PkAround put and PkDistance, i.e.,
Calculate P(k-MinPts)To P(k+MinPts)Point in range is to PkDistance.
6. pedestrian tracting method in a kind of Dancing Robot room based on laser radar according to claim 1, feature
It is, the specific method is as follows for step (3) pedestrian's identification：
(301) point in the class cluster that clusters out is scanned first, obtains the information of each class cluster, including collection point in class cluster
Number Nc, average distance D of the class cluster apart from laser radaravgAnd the position coordinates at the edge up and down of class cluster；
(302) coarse sizing is carried out according to the size of class cluster；
(303) fine screening is carried out to remaining cluster, is obtained laser radar first to the distance of the cluster, is denoted as r, enables r=
DavgIf pedestrian's waist width is Wped, WpedValue range be Wmin<Wped<Wmax；If using r as radius, by WpedAs arc length,
Then there is following formula：
The corresponding central angle θ of sector region is calculated, formula is as follows：
The number P of collection point is calculated using formula (7) according to the corresponding central angle θ of the angular resolution β and cluster of laser radarc, calculate public
Formula is as follows：
If the number of the corresponding point of pedestrian's cluster is denoted as Pnum, then PnumRange estimated value computational methods it is as follows：
To each cluster after cluster judged according to the above method, remove non-pedestrian cluster, using pedestrian's cluster as pedestrian with
The input data of track.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610439857.2A CN106125087B (en) | 2016-06-15 | 2016-06-15 | Pedestrian tracting method in Dancing Robot room based on laser radar |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610439857.2A CN106125087B (en) | 2016-06-15 | 2016-06-15 | Pedestrian tracting method in Dancing Robot room based on laser radar |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106125087A CN106125087A (en) | 2016-11-16 |
| CN106125087B true CN106125087B (en) | 2018-10-30 |
Family
ID=57469953
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610439857.2A Active CN106125087B (en) | 2016-06-15 | 2016-06-15 | Pedestrian tracting method in Dancing Robot room based on laser radar |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106125087B (en) |
Families Citing this family (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106843280B (en) * | 2017-02-17 | 2021-03-16 | 深圳市卓兴半导体科技有限公司 | Intelligent robot following system |
| EP3388864A1 (en) * | 2017-04-10 | 2018-10-17 | Bea S.A. | Method of human body recognition and human body recognition sensor |
| CN107092749A (en) * | 2017-04-22 | 2017-08-25 | 吉林大学 | Sport biomechanics coupling analysis system |
| CN107742303B (en) * | 2017-09-29 | 2021-05-25 | 南京阿凡达机器人科技有限公司 | Target image display processing method and system based on robot |
| CN107831490A (en) * | 2017-12-01 | 2018-03-23 | 南京理工大学 | A kind of improved more extension method for tracking target |
| CN108152832A (en) * | 2017-12-11 | 2018-06-12 | 开沃新能源汽车集团有限公司 | Pass through the method for laser radar detection target obstacle |
| CN108427112A (en) * | 2018-01-22 | 2018-08-21 | 南京理工大学 | A kind of improved more extension method for tracking target |
| CN110609254B (en) * | 2018-06-15 | 2022-09-27 | 富士通株式会社 | Action detection method and device based on wireless signals and electronic equipment |
| CN109543046A (en) * | 2018-11-16 | 2019-03-29 | 重庆邮电大学 | A kind of robot data interoperability Methodologies for Building Domain Ontology based on deep learning |
| CN109822563A (en) * | 2018-12-08 | 2019-05-31 | 浙江国自机器人技术有限公司 | Task follower method for IDC robot |
| CN109613553B (en) * | 2018-12-18 | 2021-05-18 | 歌尔光学科技有限公司 | Method, device and system for determining number of objects in scene based on laser radar |
| CN109934127B (en) * | 2019-02-27 | 2020-12-15 | 电子科技大学 | Pedestrian identification and tracking method based on video image and wireless signal |
| CN110210417B (en) * | 2019-06-05 | 2021-09-28 | 达闼机器人有限公司 | Method, terminal and readable storage medium for predicting pedestrian motion trail |
| CN110327187B (en) * | 2019-07-10 | 2021-07-13 | 河北工业大学 | Model-free control method with prior moment for exoskeleton |
| CN110609281B (en) * | 2019-08-23 | 2021-08-17 | 珠海格力电器股份有限公司 | Region detection method and device |
| CN111708042B (en) * | 2020-05-09 | 2023-05-02 | 汕头大学 | Robot method and system for predicting and following pedestrian track |
| CN111583257B (en) * | 2020-05-28 | 2023-07-14 | 中国铁道科学研究院集团有限公司 | Railway limit foreign matter intrusion detection method, device and system |
| CN112562419B (en) * | 2020-11-03 | 2022-04-08 | 南京航空航天大学 | Off-line multi-target tracking-based weather avoidance zone setting method |
| CN112493928B (en) * | 2020-11-26 | 2021-12-17 | 广东盈峰智能环卫科技有限公司 | Intelligent robot self-following method, device, medium and electronic equipment |
| CN113569157B (en) * | 2021-06-25 | 2023-08-29 | 深圳震有科技股份有限公司 | Radar scanning effect simulation method, system, terminal equipment and storage medium |
| CN114237264A (en) * | 2022-02-23 | 2022-03-25 | 博学宽行(成都)科技有限公司 | Robot control method based on laser and visual recognition navigation |
| CN115329899A (en) * | 2022-10-12 | 2022-11-11 | 广东电网有限责任公司中山供电局 | Clustering equivalent model construction method, system, equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101393264A (en) * | 2008-10-12 | 2009-03-25 | 北京大学 | Moving target tracking method and system based on multi-laser scanner |
| CN103325115A (en) * | 2013-06-14 | 2013-09-25 | 上海交通大学 | Pedestrian counting monitoring method based on head top camera |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2009168578A (en) * | 2008-01-15 | 2009-07-30 | Advanced Telecommunication Research Institute International | Measurement device and measurement method |
| CN102411371A (en) * | 2011-11-18 | 2012-04-11 | 浙江大学 | Multi-sensor service-based robot following system and method |
| CN103268616B (en) * | 2013-04-18 | 2015-11-25 | 北京工业大学 | The moveable robot movement human body tracing method of multi-feature multi-sensor |
-
2016
- 2016-06-15 CN CN201610439857.2A patent/CN106125087B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101393264A (en) * | 2008-10-12 | 2009-03-25 | 北京大学 | Moving target tracking method and system based on multi-laser scanner |
| CN103325115A (en) * | 2013-06-14 | 2013-09-25 | 上海交通大学 | Pedestrian counting monitoring method based on head top camera |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106125087A (en) | 2016-11-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106125087B (en) | Pedestrian tracting method in Dancing Robot room based on laser radar | |
| CN110084272B (en) | Cluster map creation method and repositioning method based on cluster map and position descriptor matching | |
| CN111665842B (en) | Indoor SLAM mapping method and system based on semantic information fusion | |
| US8824781B2 (en) | Learning-based pose estimation from depth maps | |
| CN108445480A (en) | Mobile platform based on laser radar adaptively extends Target Tracking System and method | |
| CN107907124A (en) | The localization method known again based on scene, electronic equipment, storage medium, system | |
| CN111781608B (en) | Moving target detection method and system based on FMCW laser radar | |
| CN107182036A (en) | The adaptive location fingerprint positioning method merged based on multidimensional characteristic | |
| CN110908374B (en) | Mountain orchard obstacle avoidance system and method based on ROS platform | |
| CN110379168A (en) | A kind of vehicular traffic information acquisition method based on Mask R-CNN | |
| US20230184921A1 (en) | Radar point cloud-based posture determination system | |
| CN106296743A (en) | A kind of adaptive motion method for tracking target and unmanned plane follow the tracks of system | |
| CN107728612A (en) | Identify that different crowd carries out method, storage device and the mobile terminal of advertisement pushing | |
| CN112150805B (en) | Determination method, device and equipment of travelable area and storage medium | |
| CN111709988B (en) | Method and device for determining characteristic information of object, electronic equipment and storage medium | |
| Navarro-Serment et al. | LADAR-based pedestrian detection and tracking | |
| Du et al. | A passive target recognition method based on LED lighting for industrial internet of things | |
| Razlaw et al. | Detection and tracking of small objects in sparse 3d laser range data | |
| CN106127119A (en) | Joint probabilistic data association method based on coloured image and depth image multiple features | |
| CN115808170B (en) | Indoor real-time positioning method integrating Bluetooth and video analysis | |
| Liu et al. | The detection and following of human legs based on feature optimized HDBSCAN for mobile robot | |
| CN113721633B (en) | Mobile robot path planning method based on pedestrian track prediction | |
| CN115406445A (en) | Multi-sensor data fusion processing method and robot obstacle avoidance method | |
| CN109344685A (en) | A kind of wisdom pallet and its intelligent positioning method for tracing | |
| Mohsin et al. | Localization and identification of body extremities based on data from multiple depth sensors |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |