CN102789689A - Vehicle detecting and classifying system and method - Google Patents

Abstract

The invention relates to a vehicle detecting and classifying system and method. The system comprises an image acquiring module, an image identifying module used for identifying vehicles in an image, a length computing module, and a cascading Fisher linear classifier, wherein the length computing module is used for computing the ratio of the hood to the length of the whole vehicle, the ratio is used as a first characteristic parameter ratio1, the length computing module is also used for computing the ratio of the front length to the rear length of the vehicle, and the ratio is used as a second characteristic parameter ratio2; and the cascading Fisher linear classifier is used for classifying the vehicles at least on the basis of the ratio1 and the ratio2. Preferably, the ratio of the pixel cumulative sum of the front half part of the vehicle to the pixel cumulative sum of the rear half part of the vehicle is used as a third characteristic parameter ratio3, the ratio of the pixel cumulative sum of the front left part of the vehicle to the pixel cumulative sum of the front right part of the vehicle is used as a fourth characteristic parameter ratio4, and the cascading Fisher linear classifier is also used for classifying the vehicles at least on the basis of the ratio3 and the ratio4. The system and the method disclosed by the invention are used for identifying the vehicles in the image and extracting the characteristics of the vehicles in the image, the vehicles are classified by the cascading Fisher linear classifier, and the cost is lower.

Description

Vehicle detection and categorizing system and method

[technical field]

The present invention relates to Flame Image Process and area of pattern recognition, particularly, relate to a kind of vehicle detecting system and detection method.

[background technology]

Along with the increase day by day of automobile pollution, this great social danger of road traffic accident has caused people's attention day by day.

Chinese patent 200710175796.4 provides a kind of method that vehicle is detected and classifies based on single-frequency continuous wave radar.In this method with the time domain radar signal of single-frequency continuous wave radar as input, carry out time-domain analysis and obtain time dependent Doppler frequency spectrum figure, be the parametric image of vehicle scattering center position with the Doppler frequency spectrum image mapped through hough transform.Carry out feature extraction, screening, compression then and obtain feature samples, utilize Fisher criterion gold necklace sample classification at last, obtain the vehicle classification of vehicle.This scheme implementation more complicated of getting up, cost is higher, therefore, needs a kind of improved vehicle detecting system and detection method badly.

[summary of the invention]

One aspect of the present invention provides vehicle detection and categorizing system cheaply, and this system comprises: image collection module is used to obtain image; Picture recognition module is used for the vehicle of recognition image; The length computation module is used at least the ratio based on said Image Acquisition hood and whole vehicle commander, with this ratio as first characteristic parameter; Also being used at least based on said image is the front portion of vehicle and the separatrix at rear portion with the perpendicular bisector of hood, obtains the ratio of the anterior length and the rear portion length of vehicle, with this ratio as second characteristic parameter; Cascade Fisher linear classifier is used for based on said first characteristic parameter, second characteristic parameter said vehicle being classified at least.

In a preferred version; Said vehicle detection and categorizing system also comprise the pixel computing module; Said pixel computing module is at least based on said image; Vehicle is divided into first half and latter half, the pixel accumulation of calculating said first half and latter half respectively with, with the pixel accumulation of first half and with the pixel accumulation of latter half and ratio as the 3rd characteristic parameter; Said cascade Fisher linear classifier is also classified to said vehicle based on said the 3rd characteristic parameter at least.

In a preferred version; Said pixel computing module also is divided into preceding left half and preceding right half with the first half of vehicle; Before calculating respectively the accumulation of the pixel of left half and preceding right half with, with the pixel accumulation of preceding left half and with the pixel accumulation of preceding right half and ratio as the 4th characteristic parameter; Said cascade Fisher linear classifier is also classified to said vehicle based on said the 4th characteristic parameter at least.

In a preferred version, said pixel computing module is also at least based on said image, the pixel accumulation of calculating said vehicle and whole window respectively with, with the pixel accumulation of vehicle and with the pixel accumulation of window and ratio as the 5th characteristic parameter; Said cascade Fisher linear classifier is also classified to said vehicle based on said the 5th characteristic parameter at least.

In a preferred version, said cascade Fisher linear classifier has three grades, and being respectively applied for just, vehicle is referred to car, passenger vehicle, lorry.

Another aspect of the present invention also provides a kind of vehicle detection and sorting technique, and this method may further comprise the steps: obtain image, the vehicle in the recognition image; At least based on said image, obtain hood and whole vehicle commander's ratio, with this ratio as first characteristic parameter; At least based on said image, be the front portion of vehicle and the separatrix at rear portion with the perpendicular bisector of hood, obtain the ratio of the anterior length and the rear portion length of vehicle, with this ratio as second characteristic parameter; At least based on said first characteristic parameter, second characteristic parameter, use cascade Fisher linear classifier that said vehicle is classified.

In a preferred version; Further comprising the steps of: at least based on said image; Vehicle is divided into first half and latter half; The pixel accumulation of calculating said first half and latter half respectively with, with the pixel accumulation of first half and with the pixel accumulation of latter half and ratio as the 3rd characteristic parameter; At least based on said the 3rd characteristic parameter, use cascade Fisher linear classifier that said vehicle is classified.

In a preferred version; Further comprising the steps of: that the first half of vehicle is divided into preceding left half and preceding right half; Before calculating respectively the accumulation of the pixel of left half and preceding right half with, with the pixel accumulation of preceding left half and with the pixel accumulation of preceding right half and ratio as the 4th characteristic parameter; At least based on said the 4th characteristic parameter, use cascade Fisher linear classifier that said vehicle is classified.

In a preferred version, further comprising the steps of: at least based on said image, the pixel accumulation of calculating said vehicle and whole window respectively with, with the pixel accumulation of vehicle and with the pixel accumulation of window and ratio as the 5th characteristic parameter; At least based on said the 5th characteristic parameter, use cascade Fisher linear classifier that said vehicle is classified.

In a preferred version, use three grades Fisher linear classifier that vehicle is referred to car, passenger vehicle, lorry.

The present invention discerns and feature extraction the vehicle in the image, and uses cascade Fisher linear classifier that vehicle is classified, and cost is lower.

[description of drawings]

Fig. 1 is the vehicle detection that provides of one embodiment of the invention and the framework synoptic diagram of categorizing system;

Fig. 2 is the vehicle detection that provides of one embodiment of the invention and the process flow diagram of sorting technique.

[embodiment]

With reference to figure 1 and Fig. 2, vehicle detection that one embodiment of the invention provides and categorizing system image collection module 11, picture recognition module 13, length computation module 15, pixel computing module 17 and cascade Fisher linear classifier 19.

Image collection module 11 is used to obtain image, image collection module 11 can be from storage medium reading images, also can directly take vehicle and obtain image through filming apparatus.Therefore, image collection module 11 can be a camera.

Picture recognition module 13 links to each other with image collection module 11, is used for vehicle and the hood thereof etc. (with reference to the step 21 of figure 2) of recognition image.Prior art has had several different methods from the image that includes vehicle, to identify other parts of vehicle, hood and vehicle, and therefore, the present invention is not described in detail this.

Length computation module 15 is connected with picture recognition module, is used to calculate the relevant characteristic parameter of vehicle.For example, the shared whole vehicle commander's of length of hood proportion is designated as ratio1 (with reference to the step 22 of figure 2) as first characteristic parameter.In general, the ratio1 value of car is about between the 0.2-0.5; The ratio1 value of passenger vehicle is about 1, and the hood length of coach and the ratio of whole length of wagon are also more than 0.85; The hood length of lorry and the ratio of whole length of wagon are just much little.

Again for example, be benchmark with the perpendicular bisector of hood, vehicle is divided into front and rear, the length ratio of the front and rear part of vehicle as second characteristic parameter, is designated as ratio2 (with reference to the step 25 of figure 2).For car, the ratio2 value of car is about between the 1-3, be lower than 1 more rare; For passenger vehicle, the ratio2 value is approximately about 1, and estimation is between 1-1.5.

Pixel computing module 17 is connected with picture recognition module 13.After vehicle in the image is divided into first half and latter half; The pixel accumulation that pixel computing module 17 calculates first half and latter half respectively with; With the accumulation of the pixel of first half and with the pixel accumulation of latter half and ratio as the 3rd characteristic parameter, be designated as ratio3.

Further; The first half of vehicle also is divided into preceding left half and preceding right half; Before pixel computing module 17 calculates respectively the accumulation of the pixel of left half and preceding right half with; With the accumulation of the pixel of preceding left half and with the pixel accumulation of preceding right half and ratio as the 4th characteristic parameter, be designated as ratio4.

Alternatively, at first vehicle is divided into a preceding left side, the preceding right side, right four parts in a left side, back and back according to length, then, pixel computing module 17 calculate respectively above-mentioned tetrameric pixel accumulation with, be designated as sum1, sum2, sum3 and sum4 respectively.So, ratio3=(sum1+sum2)/(sum3+sum4), ratio4=sum1/sum2 (with reference to the step 27,29 of figure 2).

Further; The pixel accumulation that pixel computing module 17 also calculates vehicle and whole window (entire image just) respectively with; With the accumulation of the pixel of vehicle and with the pixel accumulation of window and ratio as the 5th characteristic parameter, be designated as ratio5 (with reference to the step 31 of figure 2).

Cascade Fisher linear classifier 19 is connected with length computation module 15, pixel computing module 17, based on part or all of the first characteristic parameter ratio1, the second characteristic parameter ratio2, the 3rd characteristic parameter ratio3, the 4th Partial Feature parameter ratio4, the 5th characteristic parameter ratio5 to vehicle classify (with reference to the step 31 of figure 2).In the present embodiment, cascade Fisher linear classifier 17 has level Four, is respectively applied for just that vehicle is referred to car, passenger vehicle, lorry, non-vehicle.

Present embodiment is to satisfy the requirement of Fisher criterion function to gathering in the class; A large amount of rambling negative samples are carried out cluster; Form a class set to the sample that the close and distant relation of sample is closer, make its each subclass that comparatively significantly general character all arranged, and then require the interior fusion of its type to get possibility.Because negative sample is not a vehicle, possibly be various things, difference is very big, forms a class set, the accuracy in the time of just can improving linear discriminant to them through cluster analysis.

On clustering method, present embodiment adopts the K-mean cluster, is divided into 4 types to sample, i.e. car, passenger vehicle, lorry and non-vehicle to be detected.We will minimize distance in total class (sample is to the summation of the distance of corresponding particle in all kinds of).Usually, n all sample separation of search spent pretty troublesome even impossible thing of time in c cluster, only if sample number is quite little.So, our actual search be a local minimum point of distance in total type, method is to repeat to adjust the barycenter m of 4 clusters _j, and with each sample dispensing in the classification at nearest barycenter place:

$E=\underset{j=1}{\overset{c}{\mathrm{\Σ}}}\underset{x_i\∈\mathrm{\ω}/j}{\mathrm{\Σ}}{\left|\right|x_i-m_j\left|\right|}^2$

In the following formula, the error when E also can regard cluster as (summation of square deviation).

Preferably, be to have two threshold values of use to come sample is judged what differentiate.A threshold value is used for judging belonging to confirms that it belongs to positive sample, and a threshold value is used for confirming that it belongs to negative sample, for these two unascertainable samples of threshold value, judges that then they belong to intermediate sample, flows into next stage and judges.

Further, present embodiment all is designed to the sorters at different levels of cascade one two types Fisher linear classifier.Two types of sample average m have separately been obtained ₁And m ₂And sample population mean m.We can copy famous Anova statistical test to weigh the separation degree of classification, calculate " volume " of each type with respect to the population covariance matrix of their average point.In order to obtain a more concrete understanding, it is following to provide concrete computing formula:

$S_w=\underset{k=1}{\overset{2}{\mathrm{\Σ}}}\underset{x\∈C_k}{\mathrm{\Σ}}{(x-m_k)(x-m_k)}^{\′};$

S _b＝(m ₁-m ₂)(m ₁-m ₂)′。

Wherein, Sw is called within class scatter matrix, dispersion matrix between Sb type of being called.

Our target is exactly to seek an axis of orientation at feature space, makes to reach a maximal value along this within class scatter matrix, promptly makes the separability of classification reach maximum.This is equivalent to following this criterion function of maximization:

$J\left(x\right)=\frac{x^{\′}{S}_{b}x}{x^{\′}{S}_{w}x}$

The direction x of J (x) above the maximization can use following formulate:

$x=S_w^{-1}(m_1-m_2)$

X is the vector that 5 characteristic parameter ratio1, ratio2, ratio3, ratio4 and ratio5 that this patent chooses constitute in the formula.

As stated, the invention discloses a kind of vehicle detection of the Fisher linear classifier based on cascade and the system and method for car type identification.The present invention proposes a kind of implementation method of Fisher linear classifier of cascade, the notion of cascade runs through and acts on the whole training and the decision process of Fisher linear classifier, detects simultaneously and the function that realizes sorter of classifying is used more.Thereby the elder generation that is easy to make a strategic decision decision-making does not make it to get into the next stage sorter improves the efficiency of decision-making, and classifier function at different levels are independent, makes previous stages be difficult to the decision maker and is prone to become possibility with decision-making in this level, has reduced the risk of the decision-making of taking a risk.

For the person of ordinary skill of the art, under the prerequisite that does not break away from the present invention's design, 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 accompanying claims.

Claims (10)

1. vehicle detection and categorizing system is characterized in that, comprising:

Image collection module is used to obtain image;

Picture recognition module is used for the vehicle of recognition image;

The length computation module is used at least the ratio based on said Image Acquisition hood and whole vehicle commander, with this ratio as first characteristic parameter; Also being used at least based on said image is the front portion of vehicle and the separatrix at rear portion with the perpendicular bisector of hood, obtains the ratio of the anterior length and the rear portion length of vehicle, with this ratio as second characteristic parameter;

Cascade Fisher linear classifier is used for based on said first characteristic parameter, second characteristic parameter said vehicle being classified at least.

2. vehicle detection according to claim 1 and categorizing system is characterized in that:

Said vehicle detection and categorizing system also comprise the pixel computing module; Said pixel computing module is at least based on said image; Vehicle is divided into first half and latter half; The pixel accumulation of calculating said first half and latter half respectively with, with the pixel accumulation of first half and with the pixel accumulation of latter half and ratio as the 3rd characteristic parameter;

Said cascade Fisher linear classifier is also classified to said vehicle based on said the 3rd characteristic parameter at least.

3. vehicle detection according to claim 2 and categorizing system is characterized in that:

Said pixel computing module also is divided into preceding left half and preceding right half with the first half of vehicle; Before calculating respectively the accumulation of the pixel of left half and preceding right half with, with the pixel accumulation of preceding left half and with the pixel accumulation of preceding right half and ratio as the 4th characteristic parameter;

Said cascade Fisher linear classifier is also classified to said vehicle based on said the 4th characteristic parameter at least.

4. vehicle detection according to claim 1 and categorizing system is characterized in that:

Said pixel computing module is also at least based on said image, the pixel accumulation of calculating said vehicle and whole window respectively with, with the pixel accumulation of vehicle and with the pixel accumulation of window and ratio as the 5th characteristic parameter;

Said cascade Fisher linear classifier is also classified to said vehicle based on said the 5th characteristic parameter at least.

5. according to any described vehicle detection and categorizing system in the claim 1 to 4, it is characterized in that said cascade Fisher linear classifier has three grades, being respectively applied for just, vehicle is referred to car, passenger vehicle, lorry.

6. vehicle detection and sorting technique is characterized in that, may further comprise the steps:

Obtain image, the vehicle in the recognition image;

At least based on said image, obtain hood and whole vehicle commander's ratio, with this ratio as first characteristic parameter;

At least based on said image, be the front portion of vehicle and the separatrix at rear portion with the perpendicular bisector of hood, obtain the ratio of the anterior length and the rear portion length of vehicle, with this ratio as second characteristic parameter;

At least based on said first characteristic parameter, second characteristic parameter, use cascade Fisher linear classifier that said vehicle is classified.

7. vehicle detection according to claim 6 and sorting technique is characterized in that, and be further comprising the steps of:

At least based on said image, vehicle is divided into first half and latter half, the pixel accumulation of calculating said first half and latter half respectively with, with the pixel accumulation of first half and with the pixel accumulation of latter half and ratio as the 3rd characteristic parameter;

At least based on said the 3rd characteristic parameter, use cascade Fisher linear classifier that said vehicle is classified.

8. vehicle detection according to claim 7 and sorting technique is characterized in that, and be further comprising the steps of:

Left half and preceding right half before the first half of vehicle is divided into, before calculating respectively the pixel accumulation of left half and preceding right half with, accumulate with the pixel accumulation of preceding left half with the pixel of preceding right half and ratio as the 4th characteristic parameter;

At least based on said the 4th characteristic parameter, use cascade Fisher linear classifier that said vehicle is classified.

9. vehicle detection according to claim 6 and sorting technique is characterized in that, and be further comprising the steps of:

At least based on said image, the pixel accumulation of calculating said vehicle and whole window respectively with, with the pixel accumulation of vehicle and with the pixel accumulation of window and ratio as the 5th characteristic parameter;

At least based on said the 5th characteristic parameter, use cascade Fisher linear classifier that said vehicle is classified.

10. according to any described vehicle detection and sorting technique in the claim 6 to 9, it is characterized in that, use three grades Fisher linear classifier that vehicle is referred to car, passenger vehicle, lorry.

Images