CN102982802A

CN102982802A - Vehicle feature recognition algorithm based on real-time coding

Info

Publication number: CN102982802A
Application number: CN2012105162249A
Authority: CN
Inventors: 李智; 秦旭; 杨鹏
Original assignee: Sichuan University
Current assignee: Sichuan University
Priority date: 2012-12-06
Filing date: 2012-12-06
Publication date: 2013-03-20

Abstract

The invention relates to a vehicle feature recognition algorithm based on real-time coding, which is a method based on coding and feature extraction on a signal time domain. The invention provides a symbol table with 40 characters to adapt to the feature extraction of a vehicle sound signal; and the vehicle feature recognition algorithm has the characteristic of high recognition rate. Through the vehicle feature recognition algorithm, the vehicle sound signal is only coded on the time domain, and a one-dimensional matrix with fixed size is finally formed to achieve the purpose of recognizing a vehicle type. Compared with a traditional method based on frequency domain and wavelet feature extraction, the vehicle feature recognition algorithm has the beneficial effects that the operation energy and the memory resources required are less, and the implementation is easy.

Description

A kind of vehicle characteristics recognizer based on real-time coding

Technical field

The present invention proposes a kind of method of utilizing voice signal to carry out vehicle identification, it is a kind ofly to encode and the method for feature extraction in the signal time domain, the present invention is based on the symbol table of 40 characters, by the feature extraction of this symbol table to the vehicle sounds signal, be exactly a kind of method that the vehicle sounds signal is processed in brief.

Background technology

Vehicle type recognition is a very important signal processing tasks, and it can be used in the civil area such as intelligent transport system.At present external research mainly is to adopting sound transducer, vibration transducer and infrared sensor to carry out the collection of signals of vehicles and adopting various signal processing means to carry out signal and process.

Utilizing the sensor node of various miniature active or passive types to carry out the detection of signals of vehicles and processing is nowadays in the trend in vehicle identification field, and this also has higher requirement for recognizer.Traditional recognizer such as FFT are difficult to apply on the miniature sensor node, and typical 8 single-chip microcomputers are finished one 512 the about 30s of FFT computing needs under the clock frequency of 4MHz, and this can't be satisfied with the Real time identification requirement for vehicle.Also can expend a large amount of computational resources and energy for the method for utilizing wavelet transformation equally, this just becomes especially difficult so that utilize microsensor to carry out the vehicle Real time identification.

Summary of the invention

For the existing issue in background technology, the present invention proposes a kind ofly to encode and the method for feature extraction in the signal time domain, compare with traditional method based on frequency domain and Wavelet Feature Extraction, the needed computing energy of the method and memory source all seldom implement also and are easy to.

In order to reach this purpose, the present invention by the following technical solutions:

Suppose that a period of time length is T, bandwidth is the vehicle sounds signal of W, this signal comprises 2TW zero point, wherein the position feature information of zero crossing is easy to calculate, difficulty just is many but the position feature information calculations of return-to-zero point is got up, and therefore abandons calculating accurate position feature information at zero point, then signal segmentation is become a plurality of first sheets, each unit is take adjacent zero crossing as the border, and the return-to-zero dot position information in the unit just is limited in the unit like this.So just the position feature information of calculating signal zero crossing is converted into the characteristic information of Computing Meta, the characteristic information of unit is represented by two descriptors:

(1) duration D: the sampling number between adjacent zero crossing, it provides the frequency information of signal waveform,

（1）

Signal frequency,

It is sample frequency;

(2) form S: the number of maximum value or minimum value in the unit, it provides the harmonic information of signal waveform.

The result of coding is exactly the two-dimensional space that the unit that signal is all is mapped as a max (D) * max (S), but this two-dimensional space can be very large, and this depends on bandwidth and the sample frequency of signal.In order to reduce the quantity of signal description symbol, utilize a symbol table that has defined with two dimension

Descriptor is converted into the symbol stream of one dimension, and this symbol stream further is created as a size again and is

The one dimension matrix,

Be the quantity of contained symbol in the symbol table, the element in the matrix is the frequency that each symbol in the symbol table occurs in symbol stream, and we are called this matrix

Matrix, its establishment are with following expression formula:

（2）

It is of matrix

Individual element,

The quantity of symbol in the is-symbol table,

Individual unit,

All first quantity,

Expression the

The denotational description of individual unit,

（3）

When

During for 3500Hz, according to formula (1), the value of duration D is 35 to the maximum, the value of form S is limited to is 5 to the maximum, just can obtain the D/S descriptor and mostly be 35 * 5=175 most, then must utilize predefined symbol table that the D/S descriptor is converted into one dimension symbol stream, better represent the feature of lorry voice signal in order to make coding result, adopting a kind of symbol quantity is 40 symbol table.

Shown in the following form 1 of the symbol table that predefine is good:

D↓/S→	0	1	2	3	4	5
							1	1
2	1
							3	2	2
4	3	3
							5	4	5	5
6	6	5	5
							7	7	5	5	5
8	7	5	5	5
							9	8	9	9	9	9
10	8	9	10	10	10	10
							11	9	9	10	10	10	11
12	9	9	10	10	11	11
							13	10	10	11	11	11	12
14	11	11	12	12	12	13
							15	12	12	12	13	13	14
16	13	13	13	14	14	15
							17	14	14	14	15	15	16
18	15	15	15	16	16	17
							19	16	16	17	17	17	18
20	17	17	18	18	19	19
							21	18	18	19	19	20	21
22	19	19	20	20	21	22
							23	20	20	21	21	22	23
24	21	21	22	22	23	24
							25	22	22	23	24	24	25
26	23	24	24	25	25	26
							27	24	25	25	26	26	27
28	25	25	26	26	27	28
							29	26	27	27	27	28	29
30	27	28	28	29	29	30
							31	28	29	29	30	31	32
32	29	30	31	31	32	33
							33	30	31	32	32	33	34
34	34	35	36	37	38	39
							35	35	36	37	38	39	40

Form 1

Description of drawings

Accompanying drawing 1 is signal waveform segmenting principle synoptic diagram.

Accompanying drawing 2 is the principle schematic of signal description factor D and S.

Embodiment

Shown in accompanying drawing 1,2, following step is adopted in the enforcement of this coding:

(1) signal is divided into continuous some time section, each time period, each time period was referred to as unit take two of signal adjacent zero crossings as the border, and one has 8 units in the accompanying drawing 1, was respectively unit 1, unit 2, unit 3, unit 4, unit 5, unit 6, unit 7, unit 8;

(2) each first characteristic is described the factors with two and is represented: the sampling number in the duration D(elementary time section) and the number of the interior minimal value of form S(elementary time section or maximum value), in the accompanying drawing 2, the duration of D1 representative element 1, the form of S1 representative element 1, other are corresponding successively.Shown in accompanying drawing 2, D1=6, S1=0, D2=6, S2=0, D3=8, S3=0, D4=9, S4=1, D5=3, S5=0, D6=12, S6=1, D7=12, S7=1, D8=5, S8=0;

(3) utilizing the character list of form 1 D/S that each is first to encoding, each unit is represented with a character, is 6,6,7,9,2,9,9,4 with the D/S in the step 2 to the result after encoding;

(4) set up the one dimension matrix of a fixed size, this matrix comprises the probability that each symbol occurs, and different vehicle sounds signals can form different one dimension matrixes, and this matrix will be used to carry out the identification of vehicle sounds signal.

Claims

1. vehicle characteristics recognizer based on real-time coding is characterized in that: this algorithm is used for the vehicle sounds signal is carried out feature extraction and identification, and algorithm process is to carry out in the time domain of signal.

2. the algorithm for the vehicle sounds signal being carried out feature extraction and identification according to claim 1, it is characterized in that with the symbol table of predefined 40 symbols temporal signatures D and the S of vehicle sounds signal being converted into an one dimension matrix, this matrix is used for the identification of vehicle sounds signal.