CN201181487Y - Infrared opto-electric passenger traffic collection apparatus based on radial base neural network - Google Patents

Abstract

The utility model relates to an infrared photo-electronic passenger-flow acquisition device based on RBFNN (radial basis function neural network). The device combines the infrared photo-electronic detection technology with the RBFNN algorithm, establishes a passenger-flow counting system based on the RBFNN, and is suitable for the passenger-flow acquisition in public places. Four groups of correlated infrared photo-electronic sensors are arranged in the positions at the two sides of the entrance and exit of a commercial center at the height of an ankle. The sensors are connected with a computer via a switching value interface card, thereby enabling variation signals caused by customers passing by an infrared induction zone and shading the sensors to be stored in a base address in the computer. The device adopts an effective data storage structure to store the data in the base address, and to carry out a series of processing, such as pretreatment, separation, feature extraction, etc., and results are taken as the input of the RBFNN, therefore, the device can correctly identify the passenger-flow number passing by the infrared induction zone. The method improves the real-time passenger-flow counting accuracy, and can identify a plurality of persons side by side with relatively low error rate.

Description

Infrared photoelectric passenger flow collecting device based on radial base neural net

Technical field

The utility model relates to Automatic Measurement Technique, particularly a kind of infrared photoelectric passenger flow collecting device based on radial base neural net, infrared electro detection technique and radial base neural net algorithm are combined, set up a passenger flow counting system, be used for the passenger flow collection of public place based on the RBF neural network.

Background technology

Passenger flow data is significant for the industry that relies on passenger flow.For the industry of profitability, the quantity of passenger flow and resident custom have directly determined the formulation of marketing decision-making; For non-profit industry, the quantity of statistics passenger flow itself just equals to carry out safety guard work, has guaranteed the safety of statistical regions.Along with the continuous development of statistical analysis technique and computer technology, passenger flow statistics has begun to enter automatic phase, and requirement can provide immediately, reliable volume of the flow of passengers information.

Chinese patent CN200610129636 discloses a kind of infrared photoelectric passenger flow statistic device based on manikin, comprise infrared electro transmitter array and photoreceptor array, control/judging unit, have relation one to one between the transmitter of two arrays and the receiver; Described control/judging unit is made of microprocessor and peripheral circuit thereof, controlling described infrared electro transmitter array transmits, and to handling by the signal that photoreceptor array collected, learn algorithm according to manikin and distinguish people and object, whether the someone passes through in judgement, according to the order of each row's photoelectricity receiving array reception infrared light, judge the human body direction of passage; When many people passed through simultaneously, judgement was by people's number.This device is bigger to the hardware device demand, and related manikin is learned algorithm, judge according to body shape, and the algorithm complexity, therefore erroneous judgement appears in the thought of lack of wisdomization easily.

Chinese patent CN 200510060288 discloses a kind of public traffice passenger flow statistical method based on stereoscopic vision, this method utilizes processor that the binocular image that stereo vision apparatus obtains is carried out stereopsis, obtain each point in the scene to the distance between the video camera, threshold values is set on distance then, obtain apart from the scene in a certain distance range of video camera have a few, by to these denoisings, fit, again in conjunction with the characteristic recognition method of monocular image, the set of the point in the scene that those proximate compositions are round is as people's head, thereby realized head part's detection, the result's that the head part is detected position again, radius, half-tone information is transferred to track algorithm and is implemented to follow the tracks of, just can judge the direction of motion of passenger flow, thereby finish the passenger flow information statistics.This patented claim, on hardware, need to adopt equipment such as video camera, cost is higher, on software, only image has been extracted human body head information, position, radius, half-tone information are transferred to track algorithm and are implemented, thereby judge the direction of motion of passenger flow, and this kind determination methods is very high to the degree of dependence of video camera, track algorithm accuracy, be difficult to carry out technically, be difficult to the resolution that reaches higher.

Feasible passenger flow acquisition method requires that lower equipment cost is arranged, and can count relatively large continuous passenger flow, but existing passenger flow collecting device and method all can not satisfy above-mentioned requirements, and also not seeing at present has this type of than proven technique.

Summary of the invention

The utility model purpose is to provide a kind of infrared photoelectric passenger flow collecting device based on radial base neural net, can overcome the deficiency of prior art.Infrared electro detection technique and radial base neural net algorithm are combined, set up a passenger flow counting system, be used for the passenger flow collection of public place based on the RBF neural network; Low, the real-time passenger flow counting accuracy rate of cost height, error rate are lower.The utility model is a kind of infrared photoelectric passenger flow collecting device and passenger flow statistical method with theoretical and practical significance.

The infrared photoelectric passenger flow collecting device based on radial base neural net that the utility model provides comprises:

Infrared correlation photoelectric sensor, switching value interface card and computing machine.

Its middle infrared (Mid-IR) correlation photoelectric sensor will block the signal that changes according to client's walking, and described infrared photocell is 4 groups.

The switching value interface card is passed to the signal that infrared photoelectric sensor produces in the coupled computing machine, as the input of RBF neural network recognition system.

Computing machine is responsible for the training of radial base neural net, pre-service and the passenger flow number Classification and Identification and the output of data.

Correlation is installed on same the straight line between the transmitting terminal of described infrared photocell and the receiving terminal, and the infrared photocell of any transmitting terminal does not wherein influence other receiving end.When object passed through, light was blocked, and the receiving end pilot lamp is bright, and exports a high level pulse; When not having object to pass through, export a low level pulse.

The transmitting terminal emission infrared wavelength signal of infrared photoelectric sensor, when infrared ray was blocked, the receiving end pilot lamp was bright, and produced signal 1, and when infrared ray was not blocked, the receiving end pilot lamp did not work, and produced signal 0.So just, can will gather the 0-1 sequence by the switching value interface card and pass to continuous computing machine, interface card is directly connected on the host slot of computing machine.By being set, switch setting above the interface card can determine the computer address preserved.At this moment, whenever the passenger flow process, infrared sensing equipment produces corresponding signal, has stored 0,1 data in the base address.

The step that the acquisition method of the utility model provides a kind of infrared photoelectric passenger flow based on radial base neural net comprises:

Block by photoelectric tube and to produce signal the client in the infrared electro district that passes by is gathered counting, the switching value interface card is passed to coupled computing machine with the signal that infrared photoelectric sensor produces, adopt the RBF recognition method of neural network patterns, handle and gather passenger flow data, train, Classification and Identification, finish tally function;

Described training process comprises that data are obtained, the design of pre-service, feature extraction, training network parameter setting, passenger flow neural network classifier;

Described Classification and Identification be the network that will train as the basis, the data after the feature extraction are handled, comprise that data are obtained, pre-service, data are cut apart, feature extraction, the identification of passenger flow neural network classifier, statistical number of person.

The step that the acquisition method of the utility model provides a kind of infrared photoelectric passenger flow based on radial base neural net comprises:

1) select 4 groups of correlation infrared photocell transmitting terminals and receiving terminal on same straight line, place the public place to import and export both sides, to the blocking of infrared photocell, signal changes when walking by this zone by client;

2) by the switching value interface card, the variable signal that infrared photocell is produced carries out scanning collection, and is transferred to computing machine, deposits the base address in;

3) data in the base address are scanned, when data change, sample, comprise the data of storage change and the moment that changes generation;

4) the above-mentioned variable signal that collects is handled, extracted its maximum feature, as the input of RBF neural metwork training and neural network recognition system;

(1) data pre-service

The first step, the denoising process: the method for employing threshold value is removed the invalid data in the data, set the variation time-continuing process and be considered as noise removal less than 0.06ms, and the state value that it is corresponding becomes 0;

Second step, normalized: first photoelectric tube state transition is begun, be 0 constantly, deduct this initial time with other each state time corresponding and draw relative time;

(2) data are cut apart

At first search for every infrared photocell, seek the position that " 1 " occurs for the first time, the starting point of cutting apart as data, from here on, each is organized photoelectric tube and checks with identical step-length, and all photoelectric tubes are considered as cut-point when being " 0 " state in duration Δ t if be found to, and carry out the data cutting operation, setting Δ t is 0.06s, and the data set after segmenting does not have the status items of data to mend 0;

(3) feature extraction

Extract the waveform character after the above-mentioned processing is maximized, its feature mainly contains waveform relative time, pulse width, pulse interval, and wherein pulse width is the duration of state 1; Pulse interval is two times that state continued between the state 1;

5) design passenger flow neural network classifier, with the data input neural network after the feature extraction, the training network parameter, with the network that trained as the basis, to the data after the feature extraction carry out the passenger flow neural network classifier identification, obtain number.

Matrix after the described feature extraction be following shown in:

The identification of described passenger flow neural network classifier, being set at of obtaining:

(1) selection of passenger flow input layer is selected according to the problem of reality, and the input layer of infrared passenger flow statistics the inside is exactly the matrix after data characteristics is extracted;

(2) passenger flow output is every group parallel number after the packet identification, comprises the situation that a people passes through, and according to the identification that the tutor is arranged, parallel at most number is 6, with 3 outputs; Every is output as 1 greater than 0.5 the time in identification, is output as 0 less than 0.5 the time, represents 1 people, 2 people, 3 people, 4 people, 5 people and 6 people by 3 output;

(3) hidden layer selects hidden layer to adopt radial basis function as excitation function, and this radial basis function is Gaussian function (Gauss).

The utility model can overcome some problems that existing passenger flow acquisition system exists, passenger flow collecting device and statistical method in the past has marked improvement relatively, and the utility model is a kind of infrared photoelectric passenger flow collecting device and passenger flow statistical method with theoretical and practical significance.Comprise:

(1) discrimination height: by taking suitable dividing method, improved the effect of cutting apart, realized identification to continuous passenger flow.And, design special feature extracting method according to the characteristics of pulse, and at utmost refinement characterizes pulse, and test findings proves that this characteristic parameter extraction can carry out the input of classifier design.Training and learning process by neural network can be discerned the also few situation of pedestrian exactly, have also improved discrimination for the situation that many people walk abreast in addition.

(2) strong interference immunity: by the denoising in the preprocessing process, can effectively remove the noise that noise, sensor that sensor itself produces produce in installing, the noise that electromagnetic wave produces etc., thereby improve the antijamming capability of system.

(3) real-time is good: because the reaction velocity of infrared induction equipment is fast, the fast operation of while microprocessor, so can real time record and the volume of the flow of passengers passed through fast of reflection, be used for the passenger flow collection of public place, passenger flow counting accuracy rate height, error rate are lower in real time.

(4) cost is low: the infrared facility cost that the utility model adopts is low, the equipment needed thereby amount is little and require lower to hardware device.

Description of drawings

Fig. 1 the utility model connects block diagram based on the infrared photoelectric passenger flow collecting device of radial base neural net.

Fig. 2 RBF neural metwork training of the present utility model process unit block diagram.

Fig. 3 RBF neural network classification of the present utility model identifying unit block diagram.

Fig. 4 data of the present utility model are obtained process flow diagram.

Fig. 5 data pretreatment process of the present utility model figure.

Fig. 6 passenger flow data of the present utility model is cut apart process flow diagram.

Fig. 7 passenger flow data feature extraction of the present utility model process flow diagram.

Embodiment

The utility model is described with reference to the accompanying drawings as follows:

The utility model mainly is divided into two parts content: the collection of passenger flow data and the processing of passenger flow data and identification.

Provided the hardware device connection layout that passenger flow data is gathered among Fig. 1, mainly formed by infrared photoelectric sensor and switching value interface card.This system is installed in the both sides that doorway, market or gateway are positioned at ankle girth height with four groups infrared correlation type photoelectric sensors, and client is carried out data acquisition.It highly is arranged on calf, apart from the about 28cm in ground.In order to distinguish the overlapping region, improve the accuracy of scanning and be convenient to the cog region calibration, the utility model adopts the four groups of infrared electro device countings that are arranged in parallel, diameter according to calf is provided with spacing, the calf diameter is between 13cm～18cm by statistics, spacing is big more to help distinguishing lap more, but considers that excessive distance can have influence on the inconvenience of installation, is 25cm so the infrared ray spacing is set.Wherein also will keep the transmitting terminal of infrared photocell and receiving terminal on same straight line, the infrared photocell of any transmitting terminal does not wherein influence other receiving end.The transmitting terminal emission infrared wavelength signal of infrared photoelectric sensor, when infrared ray was blocked, the receiving end pilot lamp was bright, and produced signal 1, and when infrared ray was not blocked, the receiving end pilot lamp did not work, and produced signal 0.So just, can will gather the 0-1 sequence by the switching value interface card and pass to continuous computing machine, interface card is directly connected on the host slot of computing machine.By being set, switch setting above the interface card can determine the computer address preserved.At this moment, whenever the passenger flow process, infrared sensing equipment produces corresponding signal, has stored 0,1 data in the base address.Approximate number certificate to be processed that Here it is.Receiving data in the base address is moments, and the state variation that can regard as with optoelectronic switch is synchronous.

The utility model is handled the passenger flow data of being gathered and identifying is mainly finished by two parts.

First's (as shown in Figure 2) RBF network training process (learning process) comprises that data are obtained, the design of pre-service, feature extraction, training network parameter setting, passenger flow neural network classifier.

Second portion (as shown in Figure 3) RBF network class identifying (decision process), he be the network (training process such as Fig. 2) that will train as the basis, handle for the data after the feature extraction.Comprise that data are obtained, pre-service, data are cut apart, feature extraction, the identification of passenger flow neural network classifier, draw number.

The detailed process of data processing following (wherein two-part obtain, pre-service consistent) in above-mentioned two parts with characteristic extraction procedure:

1) passenger flow data obtains, as shown in Figure 4.Data are taken out from the base address, be saved in the database of computer.In order to reduce data redundancy, be when data change, to sample in the native system, promptly when being changed to state 1, or when being changed to state 0, carrying out data and preserve by state 1 by state 0.Deposit the variation of data in state[] in, the moment that changes generation deposits time[in] in.

2) carry out pre-service for the data that deposit in, as shown in Figure 5.

The first step, the denoising process.Adopt the method for threshold value to remove, in preprocessing process, set the variation time-continuing process and be considered as noise, need to remove less than 0.06m.And the state[that it is corresponding] in value become 0.

Second step, normalized.Normalized is exactly according to the relative time of gathering sample.At first carry out the data translation that the time interval does not change, with time data be initialized as 0, other each state variation time corresponding all is the relative time with respect to initial time.Way begins first luminous point state transition exactly, is 0 constantly, and other each state time corresponding deducts this initial time and draws relative time.

3) pretreated data are cut apart, as shown in Figure 6.

Because client continuously enters the market, passenger flow sample data amount is bigger, and data length does not wait.The definition of cutting apart according to data is divided into several not set of overlapping region mutually with data acquisition.Data cross the definition here is exactly to block one or several photoelectric tubes between client and the client simultaneously.What at this moment everyone data can't be simple extracts.Data are mutual continuous, interactive situations.On the contrary, data are not intersected index according to there being a space in continuous yet, can carry out under the prerequisite that be independent of each other so that data can staging treating.In fact the situation that passenger flow data intersects is exactly parallel existence, and assurance does not destroy under the situation of integrality of parallel data cuts apart data.The target of cutting apart is partitioned into data Uncrossed data exactly.

So-called " oblique line split plot design " cuts apart data in the utility model employing threshold segmentation.

This method can not cut apart the continuous passenger flow data collection of vertically cutting apart " space ", and its partition principle is as follows: it be equidistant that 4 photoelectric tubes are set in test, is suitable by 4 groups of ultrared times for a certain individual so.This is set according to setting under people's condition substantially at the uniform velocity.Like this everyone by on 4 photoelectric tube waveforms to show as width suitable, mode that like this can time difference solves this problem that the space is arranged.Should determine that also the stream of people does not have overlapping phenomenon to take place, every group of data all are independently exactly, are independent of each other, and the data of cutting apart like this are just valuable.At first search for every infrared photocell, seek the position that " 1 " occurs for the first time, as the starting point of cutting apart of data.From here on, 4 groups of photoelectric tubes are checked with identical step-length.Be considered as cut-point if be found to when 4 photoelectric tubes are " 0 " state in duration Δ t, carry out the data cutting operation.Set in the statistics of process of the test, Δ t had reasonable effect in 0.06 second.Data set after segmenting does not have the status items of data to mend 0.

4) characteristic extraction procedure, as shown in Figure 7.

Feature extraction is primarily aimed at that the formed waveform of 0-1 sequence carries out, and under the situation that many people walk abreast, produces identical pattern from many people and the guild that visually sensation might be different.The arrangement of the pulse number of these patterns, width pulse, the order of pulse all are surprising similar.But be not as broad as long, look to skip over those sequential order and other information.But these can be realized by the neural network that trains through Fig. 2 process, and carry out data qualification according to the experience that has of network.

According to data characteristic, as can be seen, data are the height pulse arrangements by 4 groups of orderly photoelectric tubes.Its feature is carried out the Useful Information that maximized extraction draws is: the time interval between waveform relative time, pulse width, the pulse.Matrix after the extraction be following shown in.

In Fig. 2, the network design process in the RBF neural metwork training process is as follows:

1) design of input layer

The selection of input layer is selected according to the problem of reality, and the input layer of infrared passenger flow statistics the inside is exactly the matrix after data characteristics is extracted, so the node number of input layer is 80.

2) design of output layer

The requirement of passenger flow output is exactly every group a parallel number after the packet identification, certainly situation about also may pass through for a people.Here according to the identification that the tutor is arranged, parallel at most number is 6.With 3 outputs.It is as follows that the expression of output corresponds to table.

Output valve is set, and every is output as 1 greater than 0.5 the time in identification, is output as 0 less than 0.5 the time.The situation of representing 1 people, 2 people, 3 people, 4 people, 5 people and 6 people by 3 output.

3) hidden layer is selected

Hidden layer adopts radial basis function as excitation function, and this radial basis function is Gaussian function (Gauss).

Handle by the sample data of the identifying among Fig. 3 after to following 5 feature extractions, recognition result is as follows:

Claims (2)

1, a kind of infrared photoelectric passenger flow collecting device based on radial base neural net is characterized in that it comprises:

Infrared correlation photoelectric sensor, switching value interface card and computing machine;

Its middle infrared (Mid-IR) correlation photoelectric sensor will block the signal that changes according to client's walking, and described infrared correlation photoelectric tube is 4 groups;

The switching value interface card is passed to the signal that infrared photoelectric sensor produces in the coupled computing machine, as the input of RBF neural network recognition system;

Computing machine is responsible for the training of radial base neural net, pre-service and the passenger flow number Classification and Identification and the output of data.

2, according to the described passenger flow collecting device of claim 1, it is characterized in that correlation is installed on same the straight line between the transmitting terminal of described infrared photocell and the receiving terminal, the infrared photocell of any transmitting terminal does not wherein influence other receiving end.

Images