CN104299417A - Vehicle identification method based on waveform detection - Google Patents

Abstract

The invention provides a vehicle identification method based on waveform detection. The vehicle identification method comprises the steps of conducting average filtering on real-time geomagnetic signals, dynamically regulating a current threshold value, calculating resultant magnetic field intensity, determining waveform feature points of a resultant magnetic field intensity curve, extracting suspected waveforms, obtaining a vehicle detection result by the adoption of multiple intermediate state machines, further processing the detection result to obtain the speed of a detected vehicle, preliminarily judging the type of the detected vehicle through a credibility method and meticulously judging and deducing the type of the detected vehicle through an Euclidean distance method. The vehicle identification method meets the requirements for low power consumption and high real-time performance of microprocessors such as a single chip microcomputer, meanwhile, the vehicle identification method can keep high robustness, adaptivity and accuracy, and the vehicle identification method is good in reliability and redundancy, solves contradictions between detection of low-speed vehicles and detection of large vehicles, is beneficial for collecting vehicle information during traffic jams and can be widely applied to vehicle detection and classification in the intelligent traffic field.

Description

Based on the vehicle identification method of wave test

[technical field]

The present invention relates to geomagnetic sensing technology field, particularly a kind of vehicle identification method based on wave test.

[background technology]

Existing geomagnetic sensor vehicle identification algorithm is meeting the low-power consumption of the microprocessors such as single-chip microcomputer, under high real-time requires, the high frequency Geomagnetic signal little to Change in Mean is insensitive, easily cause undetected to slow-moving vehicle, and there is contradiction in and detection of oversize vehicle end to end at vehicle.

Therefore, a kind of vehicle identification method based on wave test that can overcome above-mentioned deficiency is provided to seem very necessary.

[summary of the invention]

In view of this, for overcoming the deficiencies in the prior art, the invention provides one and can keep higher robustness, adaptivity and comparatively high-accuracy, reliability and the good vehicle identification method based on wave test of redundance.

For achieving the above object, the vehicle identification method based on wave test provided by the invention, is characterized in that: it comprises the steps:

A) obtain real-time raw magnetic gradient intensity three axial vector signal, carry out mean filter by moving window and filter out high frequency background noise in signal, obtain level and smooth magnetic field intensity sequence; Tested vehicle, after some geomagnetic sensor detection node, obtains real-time raw magnetic gradient intensity three axial vector signal, carries out mean filter and according to (variance) dynamic conditioning background threshold to magnetic signal in real time; If then new threshold value adopts autoregressive model prediction, if then new threshold value and previous moment threshold value remain unchanged, according to following formula:

$t_x\left(k\right)=\left\{\begin{array}{cc}{t}_x(k-1)\&times;(1-\mathrm{\&eta;})+t_x\&times;\mathrm{\&eta;}& s^2\left({\hat{x}}_i\right)<h_0\&times;n\\ t_x(k-1)& s^2\left({\hat{x}}_i\right)\&GreaterEqual;h_0\&times;n\end{array}\right.,i=k-50,...,k-1$

$t_y\left(k\right)=\left\{\begin{array}{cc}{t}_y(k-1)\&times;(1-\mathrm{\&eta;})+t_y\&times;\mathrm{\&eta;}& s^2\left({\hat{y}}_i\right)<h_0\&times;n\\ t_y(k-1)& s^2\left({\hat{y}}_i\right)\&GreaterEqual;h_0\&times;n\end{array}\right.,i=k-50,...,k-1$

$t_z\left(k\right)=\left\{\begin{array}{cc}{t}_z(k-1)\&times;(1-\mathrm{\&eta;})+t_z\&times;\mathrm{\&eta;}& s^2\left({\hat{z}}_i\right)<h_0\&times;n\\ t_z(k-1)& s^2\left({\hat{z}}_i\right)\&GreaterEqual;h_0\&times;n\end{array}\right.,i=k-50,...,k-1$

Wherein t _x(k), t _y(k), t _zk () is respectively x, y, z-axis background environment base value, and k is current time, be respectively at (k-50) to (k-1) x, y in the period, the variance of z-axis Geomagnetic signal, h ₀for without the variance of vehicle through out-of-date z-axis Geomagnetic signal, η is variance enlargement factor for upgrading weights 0.05, n, and its value is 40,

B) determine the waveform character point of resultant magnetic field intensity curve, extract doubtful waveform:

(1) utilize field signal value to calculate the value g (k) of resultant magnetic field intensity, value g (k) formula of described resultant magnetic field intensity is as follows:

$g\left(k\right)=\sqrt{{\left(\frac{\hat{x}\left(k\right)-t_x\left(k\right)}{t_x\left(k\right)}\right)}^2+{\left(\frac{\hat{y}\left(k\right)-t_y\left(k\right)}{t_y\left(k\right)}\right)}^2+{\left(\frac{\hat{z}\left(k\right)-t_z\left(k\right)}{t_z\left(k\right)}\right)}^2}$

Wherein be respectively the x after filtering, y, z three-axle magnetic field value.

(2) carry out real-time characteristic in conjunction with sectional linear fitting and locally-weighted fitting and extract synthesizing magnetic field intensity waveform with amount of compressed data, extract the characteristic that the obtains input data as waveform subsequent and state machine detecting portion:

If the resultant magnetic field intensity level of current point exceedes 105% of predetermined threshold value, then tentatively judge that current is because vehicle travels the earth magnetism waveform disturbances caused, record and be labeled as doubtful waveform W, if not, then automatically skip and again take off a bit; Under doubtful state, if the resultant magnetic field intensity level of current point is less than 85% of threshold value, then module proceeds to acknowledgement state, if now holding time of this state is more than or equal to T _threshold, then block doubtful waveform W and enter output state; After one-time detection flow process terminates, get back to initial state, the flow process before circulation;

C) many intermediatenesses machine is adopted to obtain the result of vehicle detection:

(3) testing result is processed further to the speed of a motor vehicle obtaining tested vehicle: using the input of the data of doubtful waveform W as many intermediatenesses machine, output state STATE comprises " vehicle pre-detection state ", " without car state ", " false-alarm pre-detection state " and " having car state ", wherein, preset value Ls in state machine adjusts by the velocity information of vehicle (k in formula _nowchange k into _current), β is the factor of influence 0.1, Ls of front vehicle speed to the speed of current vehicle _min, Ls _maxminimum value and the maximal value of preset value Ls respectively, Ls _minvalue is 10, Ls _maxvalue is 24, k _currentit is the slope of the resultant magnetic field waveform of Current vehicle; According to the output state STATE of state machine, obtain the net result of vehicle identification, represent with vehicle detection event sequence (0-1 sequence) d (k), d (k)=0 indicates that namely without car be when state machine exports STATE for " false-alarm pre-detection state " or " without car state ", and d (k)=1 indicates that namely car is when state machine exports STATE for " having car state ";

(4) vehicle that confidence level method tentatively judges, the careful judgement deduction of Euclidean distance method obtains tested vehicle is adopted:

Described employing Euclidean distance method is carried out disaggregated classification and first great amount of samples is inputted sorting algorithm and carry out learning, training, then to the sequence inputting algorithm of synthesis magnetic field intensity after time, amplitude two parts are normalized respectively, obtain the Euclidean distance of current sequence and each sample sequence, choose the judgement classification of classification as Current vehicle of the sample nearest with current sequence.

As a kind of preferred version, described steps A) in, obtain level and smooth magnetic field intensity sequence by moving window, utilize variance to judge whether the method dynamic conditioning present threshold value needing to adopt weighting, calculate resultant magnetic field intensity; Adopt two at a distance of the geomagnetic sensor S of L ₁, S ₂receiving real-time data, and carry out foregoing smothing filtering, calculating resultant magnetic field intensity, when certain vehicle is through surveyed area, S ₁, S ₂the time point intercepting waveform crest corresponding is t ₁, t ₂, then according to the following formulae discovery speed of a motor vehicle:

$V=\frac{L}{t_2-t_1}.$

As another preferred version, described step C) in utilize many intermediatenesses machine to obtain vehicle identification result, estimate by the velocity information of last car and Current vehicle the parameter that the speed of a motor vehicle adjusts many intermediatenesses machine; Two geomagnetic sensors are utilized to obtain speed of a motor vehicle recognition result.

As another preferred version, described step sets up evaluation and test process in (4), to waveform variances sigma ² _kwith resultant magnetic field intensity maximum amplitude G _kcarry out confidence level evaluation and test, tentatively judge large class belonging to Current vehicle according to this confidence level; Concrete evaluation and test process is as follows:

Waveform variance confidence level C ₁computing formula is

Resultant magnetic field intensity confidence level C ₂computing formula be:

Overall confidence level computing formula is

C＝wC ₁+(1-w)C ₂；

Wherein, w is waveform variance reliability weight, and its value is 0.3.

The invention has the advantages that: method of the present invention is meeting the low-power consumption of the microprocessors such as single-chip microcomputer, while high real-time requires, higher robustness, adaptivity and comparatively high-accuracy can be kept, reliability and redundance good, and the contradiction solved between slow-moving vehicle and oversize vehicle detection, be conducive to information of vehicles when gathering traffic congestion, vehicle detection and the classification in the fields such as intelligent transportation can be widely used in.

[accompanying drawing explanation]

Fig. 1 is method flow diagram of the present invention.

Fig. 2 is the process flow diagram of doubtful wave test of the present invention.

Fig. 3 is the state transition process flow diagram of many intermediatenesses machine of the present invention.

[embodiment]

In order to understand the present invention better, below with reference to accompanying drawing and instantiation, the present invention will be described in detail.

Vehicle identification method based on wave test of the present invention, obtains real-time raw magnetic gradient intensity three axial vector signal, carries out mean filter and filters out high frequency background noise in signal, obtain level and smooth magnetic field intensity sequence by moving window.

According to the variance dynamic conditioning background threshold of the data obtained after mean filter, overcome background value drifting problem, if then new threshold value just adopts autoregressive model prediction, if then new threshold value and previous moment threshold value remain unchanged, according to following formula:

$t_x\left(k\right)=\left\{\begin{array}{cc}{t}_x(k-1)\&times;(1-\mathrm{\&eta;})+t_x\&times;\mathrm{\&eta;}& s^2\left({\hat{x}}_i\right)<h_0\&times;n\\ t_x(k-1)& s^2\left({\hat{x}}_i\right)\&GreaterEqual;h_0\&times;n\end{array}\right.,i=k-50,...,k-1$

$t_y\left(k\right)=\left\{\begin{array}{cc}{t}_y(k-1)\&times;(1-\mathrm{\&eta;})+t_y\&times;\mathrm{\&eta;}& s^2\left({\hat{y}}_i\right)<h_0\&times;n\\ t_y(k-1)& s^2\left({\hat{y}}_i\right)\&GreaterEqual;h_0\&times;n\end{array}\right.,i=k-50,...,k-1$

$t_z\left(k\right)=\left\{\begin{array}{cc}{t}_z(k-1)\&times;(1-\mathrm{\&eta;})+t_z\&times;\mathrm{\&eta;}& s^2\left({\hat{z}}_i\right)<h_0\&times;n\\ t_z(k-1)& s^2\left({\hat{z}}_i\right)\&GreaterEqual;h_0\&times;n\end{array}\right.,i=k-50,...,k-1$

Wherein t _x(k), t _y(k), t _zk () is respectively x, y, z-axis background environment base value, and k is current time, be respectively (k-50) to (k-1) x, y in the period, the variance of z-axis Geomagnetic signal, h ₀for without the variance of vehicle through out-of-date z-axis Geomagnetic signal, η is that to upgrade weights 0.05, n be the enlargement factor of variance, and its value is 40.

Field signal value is utilized to calculate the value g (k) of resultant magnetic field intensity according to following formula:

$g\left(k\right)=\sqrt{{\left(\frac{\hat{x}\left(k\right)-t_x\left(k\right)}{t_x\left(k\right)}\right)}^2+{\left(\frac{\hat{y}\left(k\right)-t_y\left(k\right)}{t_y\left(k\right)}\right)}^2+{\left(\frac{\hat{z}\left(k\right)-t_z\left(k\right)}{t_z\left(k\right)}\right)}^2}$

Wherein be respectively the x after filtering, y, z three-axle magnetic field value.

Carry out real-time characteristic in conjunction with sectional linear fitting and locally-weighted fitting and extract synthesizing magnetic field intensity waveform with amount of compressed data, extract the characteristic that the obtains input data as waveform subsequent and state machine detecting portion, if the resultant magnetic field intensity level of current point exceedes 105% of predetermined threshold value, then tentatively judge that current is because vehicle travels the earth magnetism waveform disturbances caused, record and be labeled as doubtful waveform, if not, then automatically skip and again take off a bit.Under doubtful state, if the resultant magnetic field intensity level of current point is less than 85% of threshold value, then module proceeds to acknowledgement state, if now holding time of this state is more than or equal to T _threshold, then block waveform and enter output state.After one-time detection flow process terminates, get back to initial state, the flow process before circulation.The process flow diagram of doubtful wave test is as Fig. 2.

Using the input of doubtful Wave data as many intermediatenesses machine, output state STATE comprises " vehicle pre-detection state ", " without car state ", " false-alarm pre-detection state " and " having car state ", and the state transition process flow diagram of its state machine is as Fig. 3;

Wherein, the preset value Ls in state machine adjusts by the velocity information of vehicle, β is the factor of influence 0.1, Ls of front vehicle speed to the speed of current vehicle _min, Ls _maxminimum value 10 and the maximal value 24, k of preset value Ls respectively _nowit is the slope of the resultant magnetic field waveform of Current vehicle.

According to the output state STATE of state machine, obtain the net result of vehicle identification, represent with vehicle detection event sequence (0-1 sequence) d (k), d (k)=0 indicates that, without car (when state machine exports STATE for " false-alarm pre-detection state " or " without car state "), d (k)=1 indicates car (when state machine exports STATE for " having car state ").

Adopt two at a distance of the geomagnetic sensor S of L ₁, S ₂receiving real-time data, and carry out foregoing smothing filtering, calculating resultant magnetic field intensity, when certain vehicle is through surveyed area, S ₁, S ₂the time point intercepting waveform crest corresponding is t ₁, t ₂, then according to the following formulae discovery speed of a motor vehicle:

$V=\frac{L}{t_2-t_1}.$

Set up evaluation and test process, to waveform variances sigma ² _kwith resultant magnetic field intensity maximum amplitude G _kcarry out confidence level evaluation and test, tentatively judge large class belonging to Current vehicle according to this confidence level.Concrete evaluation and test process is as follows:

Waveform variance confidence level C ₁computing formula is

In like manner can be derived from resultant magnetic field intensity confidence level C ₂computing formula be:

Overall confidence level computing formula is

C＝wC ₁+(1-w)C ₂；

Wherein, w is waveform variance reliability weight 0.3.

Above-mentioned employing Euclidean distance method carries out disaggregated classification: first great amount of samples is inputted sorting algorithm and carry out learning, training, then to the sequence inputting algorithm of synthesis magnetic field intensity after time, amplitude two parts are normalized respectively, obtain the Euclidean distance of current sequence and each sample sequence, choose the judgement classification of classification as Current vehicle of the sample nearest with current sequence.

Principle of work of the present invention is as follows: the vehicle identification method that the present invention is based on wave test, obtain real-time raw magnetic gradient intensity three axial vector signal and carry out mean filter by moving window and filter out high frequency background noise in signal, obtain level and smooth magnetic field intensity sequence, utilize historical data variance to follow the trail of the change of Geomagnetic signal, and judge whether the method dynamic conditioning present threshold value needing employing weighting accordingly; Resultant magnetic field intensity is calculated according to field signal value; Automatic detection background threshold value, if the resultant magnetic field intensity level of current point exceedes predetermined threshold value, then tentatively judge that current is because vehicle travels the earth magnetism waveform disturbances caused, record and be labeled as doubtful waveform, confirm that waveform terminates under doubtful state after, then block waveform and enter output state; Using the input of doubtful Wave data as many intermediatenesses machine, according to the output state STATE of state machine, obtain the net result of vehicle identification; In surveyed area, be parallel to road place two geomagnetism detecting devices, calculate speed of a motor vehicle recognition result according to the distance between two the waveform sensor ripple peak-to-peak mistimings obtained and two sensors; Set up evaluation and test process, confidence level evaluation and test is carried out to waveform variance and resultant magnetic field intensity maximum amplitude, tentatively judge large class belonging to Current vehicle according to this confidence level, then adopt Euclidean distance method to carry out disaggregated classification.First great amount of samples is inputted sorting algorithm to carry out learning, training, then to the sequence inputting algorithm of synthesis magnetic field intensity after time, amplitude two parts are normalized respectively, obtain the Euclidean distance of current sequence and each sample sequence, choose the judgement classification of classification as Current vehicle of the sample nearest with current sequence.

For overcoming threshold drift problem, background threshold according to the variance dynamic conditioning of the data obtained after mean filter, and adopts autoregressive model to predict.

For overcoming the detection contradictory problems of the undetected and oversize vehicle end to end with vehicle of slow-moving vehicle, estimate by the velocity information of last car and Current vehicle the parameter that the speed of a motor vehicle adjusts many intermediatenesses machine.

The above exemplifying embodiment only have expressed some embodiments of the present invention, and it describes comparatively concrete and detailed, but therefore can not be interpreted as the restriction to the scope of the claims of the present invention.It should be pointed out that for the person of ordinary skill of the art, without departing from the inventive concept of the premise, can also make some distortion and improvement, these all belong to protection scope of the present invention.Therefore, the protection domain of patent of the present invention should be as the criterion with claims.

Claims (4)

1. based on a vehicle identification method for wave test, it is characterized in that: it comprises the steps:

A) obtain real-time raw magnetic gradient intensity three axial vector signal, carry out mean filter by moving window and filter out high frequency background noise in signal, obtain level and smooth magnetic field intensity sequence; Tested vehicle, after some geomagnetic sensor detection node, obtains real-time raw magnetic gradient intensity three axial vector signal, carries out mean filter and according to variance dynamic conditioning background threshold to magnetic signal in real time; If then new threshold value adopts autoregressive model prediction, if then new threshold value and previous moment threshold value remain unchanged, according to following formula:

$t_x\left(k\right)=\left\{\begin{array}{cc}{t}_x(k-1)\&times;(1-\mathrm{\&eta;})+t_x\&times;\mathrm{\&eta;}& s^2\left({\hat{x}}_i\right)<h_0\&times;n\\ t_x(k-1)& s^2\left({\hat{x}}_i\right)\&GreaterEqual;h_0\&times;n\end{array}\right.,i=k-50,...,k-1$

$t_y\left(k\right)=\left\{\begin{array}{cc}{t}_y(k-1)\&times;(1-\mathrm{\&eta;})+t_y\&times;\mathrm{\&eta;}& s^2\left({\hat{y}}_i\right)<h_0\&times;n\\ t_y(k-1)& s^2\left({\hat{y}}_i\right)\&GreaterEqual;h_0\&times;n\end{array}\right.,i=k-50,...,k-1$

$t_z\left(k\right)=\left\{\begin{array}{cc}{t}_z(k-1)\&times;(1-\mathrm{\&eta;})+t_z\&times;\mathrm{\&eta;}& s^2\left({\hat{z}}_i\right)<h_0\&times;n\\ t_z(k-1)& s^2\left({\hat{z}}_i\right)\&GreaterEqual;h_0\&times;n\end{array}\right.,i=k-50,...,k-1$

Wherein t _x(k), t _y(k), t _zk () is x, y, z-axis background environment base value, and k is current time, be respectively (k-50) to (k-1) variance of x, y, z axle Geomagnetic signal in the period, h ₀for without the variance of vehicle through out-of-date z-axis Geomagnetic signal, η is that to upgrade weights 0.05, n be the enlargement factor of variance, and its value is 40;

B) determine the waveform character point of resultant magnetic field intensity curve, extract doubtful waveform:

(1) utilize field signal value to calculate the value g (k) of resultant magnetic field intensity, value g (k) formula of described resultant magnetic field intensity is as follows:

$g\left(k\right)=\sqrt{{\left(\frac{\hat{x}\left(k\right)-t_x\left(k\right)}{t_x\left(k\right)}\right)}^2+{\left(\frac{\hat{y}\left(k\right)-t_y\left(k\right)}{t_y\left(k\right)}\right)}^2+{\left(\frac{\hat{z}\left(k\right)-t_z\left(k\right)}{t_z}\right)}^2}$

Wherein be respectively the x after filtering, y, z three-axle magnetic field value;

(2) carry out real-time characteristic in conjunction with sectional linear fitting and locally-weighted fitting and extract synthesizing magnetic field intensity waveform with amount of compressed data, extract the characteristic that the obtains input data as waveform subsequent and state machine detecting portion:

If the resultant magnetic field intensity level of current point exceedes 105% of predetermined threshold value, then tentatively judge that current is because vehicle travels the earth magnetism waveform disturbances caused, record and be labeled as doubtful waveform W, if not, then automatically skip and again take off a bit; Under doubtful state, if the resultant magnetic field intensity level of current point is less than 85% of threshold value, then module proceeds to acknowledgement state, if now holding time of this state is more than or equal to T _threshold, then block doubtful waveform W and enter output state; After one-time detection flow process terminates, get back to initial state, the flow process before circulation;

C) many intermediatenesses machine is adopted to obtain the result of vehicle detection:

(3) testing result is processed further to the speed of a motor vehicle obtaining tested vehicle: using the input of the data of doubtful waveform W as many intermediatenesses machine, output state STATE comprises " vehicle pre-detection state ", " without car state ", " false-alarm pre-detection state " and " having car state ", wherein, preset value Ls in state machine adjusts, according to formula by the velocity information of vehicle wherein β is the factor of influence 0.1, Ls of front vehicle speed to the speed of current vehicle _min, Ls _maxminimum value 10 and the maximal value 24, k of preset value Ls respectively _currentit is the slope of the resultant magnetic field waveform of Current vehicle; According to the output state STATE of state machine, obtain the net result of vehicle identification, represent with vehicle detection event sequence (0-1 sequence) d (k), d (k)=0 indicates that namely without car be when state machine exports STATE for " false-alarm pre-detection state " or " without car state ", and d (k)=1 indicates that namely car is when state machine exports STATE for " having car state ";

(4) vehicle that confidence level method tentatively judges, the careful judgement deduction of Euclidean distance method obtains tested vehicle is adopted:

Described employing Euclidean distance method is carried out disaggregated classification and first great amount of samples is inputted sorting algorithm and carry out learning, training, then to the sequence inputting algorithm of synthesis magnetic field intensity after time, amplitude two parts are normalized respectively, obtain the Euclidean distance of current sequence and each sample sequence, choose the judgement classification of classification as Current vehicle of the sample nearest with current sequence.

2. the vehicle identification method based on wave test according to claim 1, it is characterized in that: described steps A) in, level and smooth magnetic field intensity sequence is obtained by moving window, utilize variance to judge whether the method dynamic conditioning present threshold value needing to adopt weighting, calculate resultant magnetic field intensity; Adopt two at a distance of the geomagnetic sensor S of L ₁, S ₂receiving real-time data, and carry out foregoing smothing filtering, calculating resultant magnetic field intensity, when certain vehicle is through surveyed area, S ₁, S ₂the time point intercepting waveform crest corresponding is t ₁, t ₂, then according to the following formulae discovery speed of a motor vehicle:

$V=\frac{L}{t_2-t_1}.$

3. the vehicle identification method based on wave test according to claim 1, it is characterized in that: described step C) in utilize many intermediatenesses machine to obtain vehicle identification result, estimate by the velocity information of last car and Current vehicle the parameter that the speed of a motor vehicle adjusts many intermediatenesses machine; Two geomagnetic sensors are utilized to obtain speed of a motor vehicle recognition result.

4. the vehicle identification method based on wave test according to claim 1, is characterized in that: described step sets up evaluation and test process in (4), to waveform variances sigma ² _kwith resultant magnetic field intensity maximum amplitude G _kcarry out confidence level evaluation and test, tentatively judge large class belonging to Current vehicle according to this confidence level; Concrete evaluation and test process is as follows:

Waveform variance confidence level C ₁computing formula is

Resultant magnetic field intensity confidence level C ₂computing formula be

overall confidence level computing formula is

C＝wC ₁+(1-w)C ₂；

Wherein, w is waveform variance reliability weight, and value is 0.3.