CN106407887B - Method and device for obtaining step size of candidate frame search - Google Patents

Abstract

The invention discloses a method for acquiring the search step length of a candidate frame, which relates to the field of target detection in computer vision and pattern recognition, and can determine the search step length of a target candidate frame in a video to be detected by establishing a Poisson distribution model. The method comprises the steps of obtaining an image to be searched; acquiring image information of each original partition in the image to be searched and a Poisson distribution function corresponding to each original partition; determining the partition type of each original partition according to the image information and the Poisson distribution function of each original partition; and determining the candidate frame searching step length in each original partition according to the partition type. The technical scheme provided by the embodiment of the invention can be suitable for target detection processes such as pedestrian detection, vehicle detection and the like in scenes such as static monitoring videos, vehicle monitoring videos and the like.

Description

Method and device for obtaining step size of candidate frame search

【Technical field】

The invention relates to the field of target detection in computer vision and pattern recognition, in particular to a method and device for obtaining a candidate frame search step size.

【Background technique】

With the rapid development and wide application of computer image processing technology, the demand for target detection technology is gradually rising. Object detection has become a basic problem in the field of computer vision and pattern recognition, and the determination of the search step size of candidate boxes for detecting objects is an important preliminary work in object recognition and classification. At present, the existing method for generating target candidate frames is generally a sliding window search method. When a target search method is performed in the sliding window search method, the candidate frame steps by a fixed length in the entire scanning window.

In the process of realizing the present invention, the inventor found that there are at least the following problems in the prior art:

According to the existing target search method, in the process of searching for the target, the candidate frame is stepped by a fixed length in the entire scanning window, and the search by a fixed length in the area with different number of search targets may appear. Missing detection, the search results are not globally optimal.

[Content of the invention]

In view of this, the embodiments of the present invention provide a method and device for obtaining a search step size of a candidate frame, which can determine the search step size of a candidate frame according to the frequency and density information of search targets appearing in an area.

On the one hand, an embodiment of the present invention provides a method for obtaining a search step size of a candidate frame, the method comprising:

Get the image to be searched;

Obtain the image information of each original partition in the to-be-searched image and the Poisson distribution function corresponding to each original partition;

Determine the partition type of each original partition according to the image information and Poisson distribution function of each original partition;

According to the partition type, the candidate box search step size in each original partition is determined.

On the other hand, an embodiment of the present invention provides a device for obtaining a search step size of a candidate frame, and the device includes:

a first acquisition unit, used to acquire the image to be searched;

a second acquiring unit, configured to acquire the image information of each original partition in the image to be searched and the Poisson distribution function corresponding to each original partition;

a first determining unit, configured to determine the partition type of each original partition according to the image information and Poisson distribution function of each original partition;

The second determining unit is configured to determine, according to the partition type, the search step size of the candidate frame in each original partition.

The embodiment of the present invention provides a method and device for obtaining a search step size of a candidate frame. By establishing a Poisson model by partition and continuously learning and updating the position information of a specific search target, the candidate frame of each area in each frame of image can be adjusted. The search step size constrains the number of target candidate boxes, so as to detect the search target for different regions. The method and device improve the detection effect, and can obtain a higher detection rate.

【Description of drawings】

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without any creative effort.

1 is a flowchart of a method for obtaining a candidate frame search step size provided by an embodiment of the present invention;

2 is a flowchart of another method for obtaining a candidate frame search step size provided by an embodiment of the present invention;

3 is a flowchart of another method for obtaining a candidate frame search step size provided by an embodiment of the present invention;

4 is a flowchart of another method for obtaining a candidate frame search step size provided by an embodiment of the present invention;

5 is a block diagram of an apparatus for obtaining a candidate frame search step size provided by an embodiment of the present invention;

6 is a block diagram of an apparatus for obtaining another candidate frame search step size provided by an embodiment of the present invention;

FIG. 7 is a block diagram of an apparatus for obtaining another candidate frame search step size provided by an embodiment of the present invention;

FIG. 8 is a block diagram of another device for obtaining a search step size of a candidate frame provided by an embodiment of the present invention.

【Detailed ways】

In order to better understand the technical solutions of the present invention, the embodiments of the present invention are described in detail below with reference to the accompanying drawings.

It should be understood that the described embodiments are only some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. As used in the embodiments of the present invention and the appended claims, the singular forms "a," "the," and "the" are intended to include the plural forms as well, unless the context clearly dictates otherwise.

It should be understood that the term "and/or" used in this document is only an association relationship to describe the associated objects, indicating that there may be three kinds of relationships, for example, A and/or B, which may indicate that A exists alone, and A and B exist at the same time. B, there are three cases of B alone. In addition, the character "/" in this document generally indicates that the related objects are an "or" relationship.

Depending on the context, the word "if" as used herein can be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detection". Similarly, the phrases "if determined" or "if detected (the stated condition or event)" can be interpreted as "when determined" or "in response to determining" or "when detected (the stated condition or event)," depending on the context )" or "in response to detection (a stated condition or event)".

An embodiment of the present invention provides a method for obtaining a candidate frame search step size, which can be applied to target detection processes such as pedestrian detection and vehicle detection in scenes including static surveillance video and vehicle surveillance video. As shown in FIG. 1 , the Methods include:

101. Acquire an image to be searched.

Wherein, the to-be-searched images refer to all the to-be-detected images in the target detection process.

102. Acquire image information of each original partition in the to-be-searched image and a Poisson distribution function corresponding to each original partition.

Among them, it should be noted that the embodiment of the present invention is aimed at scene monitoring such as static monitoring video and vehicle monitoring video, and the frequency and distribution density of the search target appearing in a certain area in the image obey Poisson distribution.

Wherein, each original partition refers to each area after the detection area is divided into blocks.

Wherein, the image information of each original partition includes frequency information of the search target appearing in each area and density information of the distribution of the search target.

The search target refers to objects to be detected in the target detection process, such as people, vehicles, and objects.

The Poisson distribution function is a function established based on a mathematical Poisson model, and is suitable for describing the number of random events occurring per unit time.

103. Determine the partition type of each original partition according to the image information of each original partition and the Poisson distribution function.

Wherein, the partition type is determined by the number of search targets in the original partition.

The number of search targets refers to the number of search target distributions per unit time, and is related to the frequency information of the search targets and the density information of the distribution.

104. Determine, according to the partition type, a search step size of the candidate frame in each original partition.

A method for obtaining a candidate frame search step size provided by an embodiment of the present invention can adjust the candidate frame search step size of each area in each frame of image by establishing a Poisson model by partition and continuously learning and updating the position information of a specific target. , which constrains the number of target candidate boxes to detect targets for different regions. This method improves the detection effect and can obtain a higher detection rate.

Further, in combination with the foregoing method flow, in another possible implementation manner of the embodiment of the present invention, for step 103, according to the image information and Poisson distribution function of each original partition, determine the realization of the partition type of each original partition The following specific processes are provided, as shown in Figure 2, including:

201. Determine the number of search targets in each original partition according to the image information of each original partition and the Poisson distribution function.

Among them, the Poisson distribution function has two parameters in the continuous distribution, one is the frequency of random events in the time domain, and the other is the distribution density of random events in the space domain. In this embodiment of the present invention, the time The frequency of occurrence of random events on the domain refers to the frequency of the search target appearing in each original partition, and the distribution density of random events on the spatial domain refers to the density of the search target distribution.

The Poisson distribution function is trained in the process of detecting the search target, and dynamically obtains frequency information of the search target in each area and distribution density information of the search target during the learning process.

Among them, when the search target is detected, each frame of the image to be searched is detected, the Poisson distribution function becomes a discrete distribution, and the time is determined, and the number of search targets is the density of the search target distribution per unit time.

202. When the number of the search targets is within the first threshold range, determine the partition type of the original partition where the number of the search targets is within the first threshold range as the target sparse distribution area.

Wherein, the first threshold range refers to a number range in which the number of search targets is relatively small, and is denoted as [L, Lmin).

Wherein, the L is an integer greater than 0.

Wherein, the Lmin is an integer greater than L.

203. When the number of the search targets is within the second threshold range, determine the partition type of the original partition with the number of the search targets within the second threshold range as the target medium distribution area.

The second threshold range refers to a number range in which the number of search targets is greater than the first threshold range, which is denoted as [Lmin, Lmax].

Wherein, the Lmax is an integer greater than Lmin.

204. When the number of the search targets is within the third threshold range, determine the partition type of the original partition with the number of the search targets within the third threshold range as the target dense distribution area.

Wherein, the third threshold range refers to a number range in which the number of search targets is greater than the second threshold range, denoted as [Lmax, +∝).

Wherein, when the number of search targets in the original partition is greater than Lmin, the original partition is regarded as an area with more targets.

Wherein, when the number of search targets in the original partition is less than L, the original partition is regarded as an area with fewer targets.

Among them, the partition type is determined for the area with more search targets, and the search step size of the candidate frame is directly determined according to the number of search targets for the area with few search targets.

Further, in combination with the foregoing method flow, another possible implementation manner of the embodiment of the present invention provides specific steps of how to determine the search step size of candidate boxes in each original partition according to the partition type. The implementation of 104 provides the following specific processes, as shown in FIG. 3 , including:

301. Set the search step size of the candidate frame in the original partition where the partition type is the target sparse distribution area to the first step size.

Wherein, the length of the first step can be defined by itself according to the distribution of the search targets in the target sparse distribution area.

302. Set the search step size of the candidate frame in the original partition where the partition type is the target medium distribution area as the second step size.

Wherein, the length of the second step can be defined by itself according to the distribution of the search targets in the target medium distribution area.

Wherein, the second step size is smaller than the first step size.

303. Set the search step size of the candidate frame in the original partition where the partition type is the target dense distribution area to the third step size.

Wherein, the length of the third step can be defined by itself according to the distribution of the search targets in the dense target distribution area.

Wherein, the third step size is smaller than the second step size.

It should be noted that, when the original partition belongs to an area with few search targets, the step size is directly determined according to the number of search targets, and the step size is not less than the first step size.

Further, in combination with the foregoing method flow, the embodiment of the present invention provides another possible implementation manner. As shown in FIG. 4 , before acquiring the image to be searched, the method further includes:

401. Obtain the original partition in the original image.

Wherein, the original image refers to n frames of images in the detection area.

Wherein, the n is an integer greater than 0.

Wherein, the original partition is divided into blocks according to the size and characteristics of the detection area.

Wherein, when the number of the original partitions is larger, the function can more accurately reflect the frequency and distribution density information of the targets in the partitions, and the statistical distribution inside each original partition can be considered to be basically the same.

Wherein, the larger the number of the original partitions, the more data needs to be calculated. Considering the complexity and accuracy of the calculation method, the number of the original partitions is determined according to the size and characteristics of the search area.

402. Collect the occurrence frequency and distribution density of the search targets in the original partition.

The collecting the occurrence frequency and distribution density of the search targets in the original partition refers to the occurrence frequency and distribution density of each original partition search target collected from the n frames of images.

403. Determine a Poisson distribution function of the original partition according to the occurrence frequency and density of the search target.

Wherein, the initial value of the Poisson distribution function is obtained through parameter estimation by collecting information on the occurrence frequency and density of the search target in the n frames of images.

A method for obtaining a candidate frame search step size provided by an embodiment of the present invention can adjust the candidate frame search step of each area in each frame of image by establishing a Poisson model by partition and continuously learning and updating the position information of a specific search target. long, constrain the number of target candidate boxes, so as to detect the search target for different regions. This method improves the detection effect and can obtain a higher detection rate.

In order to better understand the technical solution, the embodiments of the present invention provide more specific implementations, and the embodiments of the present invention that need to be explained are applicable to but not limited to the following implementations.

Step 1. Divide the detection area into 12 (3x4) small areas, establish the Poisson position of pedestrians for each area of the first 2400 frames of images, and perform parameter estimation to obtain the initial value of the model, so that the Poisson model unit at each area can be obtained. The initial value of the average incidence of pedestrian appearances over time.

Step 2: Train the Poisson model, and dynamically obtain the frequency information of pedestrians appearing in each area and the density information of pedestrian distribution during the learning process.

Step 21: According to the initial value of the established Poisson model, use the gradient descent method to learn and update to obtain the density information of each original partition distribution of pedestrians in each frame of image.

Step 22. Regarding the distribution of pedestrians in each location area obtained in step 21, for the scene in the embodiment, there are more sub-block areas for pedestrians:

When the number of pedestrians appearing is less than 10, the area is regarded as a sparse distribution area of pedestrians;

When the number of pedestrians appearing is 10 to 30, the area is regarded as a medium distribution area of pedestrians;

When the number of pedestrians is more than 30, the area is regarded as a densely distributed area of pedestrians.

Step 23: For the segmented area where there are few pedestrians, set the threshold for the number of pedestrians to be 8.

Step 3: According to the density information of pedestrian distribution in each frame image obtained by the Poisson position model, set different search step sizes of candidate boxes in the regions of different distribution density sets of the search target, and control the number of candidate boxes in the whole frame image.

Step 31: Set the step size for the area with more search targets as follows: step=16 in the area with sparse pedestrian distribution, step=8 in the area with moderate distribution of pedestrians, and step=4 in the area with dense pedestrian distribution.

Step 32 , set the step size for the area with less distribution of search targets as follows: for the case where the number of pedestrians appearing is greater than 8, the step size step=16, otherwise, the step size step=32 is set.

In the embodiment of the present invention, the aspect ratio of the candidate frame set in each area is locked to 0.5, and in addition, in the case where there is no pedestrian in each area, no candidate frame is generated.

A method for obtaining a candidate frame search step size provided by an embodiment of the present invention can adjust the candidate frame search step of each area in each frame of image by establishing a Poisson model by partition and continuously learning and updating the position information of a specific search target. long, constrain the number of target candidate boxes, so as to detect the search target for different regions. This method improves the detection effect and can obtain a higher detection rate.

An embodiment of the present invention provides a device for obtaining a search step size of a candidate frame, which can be used to implement the flow of the aforementioned methods. The composition is shown in FIG. 5 , and the device includes:

The first acquiring unit 51 is used to acquire the image to be searched.

The second acquiring unit 52 is configured to acquire the image information of each original partition in the image to be searched and the Poisson distribution function corresponding to each original partition.

The first determining unit 53 is configured to determine the partition type of each original partition according to the image information of each original partition and the Poisson distribution function.

The second determining unit 54 is configured to determine, according to the partition type, the search step size of the candidate frame in each original partition.

Optionally, as shown in FIG. 6 , the first determining unit 53 includes:

The first determination module 531 is configured to determine the number of search targets in each original partition according to the image information of each original partition and the Poisson distribution function.

The second determining module 532 is configured to, when the number of the search targets is within the first threshold range, determine the partition type of the original partition with the number of search targets within the first threshold range as the target sparse distribution area.

The third determining module 533 is configured to, when the number of the search targets is within the second threshold range, determine the partition type of the original partition with the number of search targets within the second threshold range as the target medium distribution area.

The fourth determination module 534 is configured to, when the number of the search targets is within the third threshold range, determine the partition type of the original partition with the number of search targets within the third threshold range as the target dense distribution area.

Optionally, as shown in FIG. 7 , the second determining unit 54 includes:

The first setting module 541 is configured to set the search step size of the candidate frame in the original partition where the partition type is the target sparse distribution area to the first step size.

The second setting module 542 is configured to set the search step size of the candidate frame in the original partition where the partition type is the target medium distribution area to the second step size.

The third setting module 543 is configured to set the search step size of the candidate frame in the original partition where the partition type is the target dense distribution area to the third step size.

Optionally, as shown in Figure 8, the device further includes:

The third acquiring unit 55 is configured to acquire the original partition in the original image.

The collecting unit 56 is configured to collect the occurrence frequency and distribution density of the search targets in the original partition.

The third determining unit 57 is configured to determine the Poisson distribution function of the original partition according to the occurrence frequency and density of the search target.

An apparatus for obtaining a search step size of a candidate frame provided by an embodiment of the present invention can adjust the search step of a candidate frame for each area in each frame of images by establishing a Poisson model by partition and continuously learning and updating the position information of a specific search target. long, constrain the number of target candidate boxes, so as to detect the search target for different regions. This device improves the detection effect and can obtain a higher detection rate.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above may refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

In the several embodiments provided by the present invention, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined. Either it can be integrated into another system, or some features can be omitted, or not implemented. On the other hand, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, and may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware, or may be implemented in the form of hardware plus software functional units.

The above-mentioned integrated units implemented in the form of software functional units can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium, and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor (Processor) to execute the methods described in the various embodiments of the present invention. some steps. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk and other media that can store program codes .

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the present invention. within the scope of protection.

Claims (4)

1. an acquisition method of candidate frame search step size, is characterized in that, comprises: Get the image to be searched; Obtain the image information of each original partition in the to-be-searched image and the Poisson distribution function corresponding to each original partition; Determine the partition type of each original partition according to the image information and Poisson distribution function of each original partition; According to the partition type, determine the candidate frame search step size in each original partition; Determining the partition type of each original partition according to the image information and Poisson distribution function of each original partition includes: Determine the number of search targets in each original partition according to the image information and Poisson distribution function of each original partition; When the number of the search targets is within the first threshold range, the partition type of the original partition with the number of search targets within the first threshold range is determined as the target sparse distribution area; When the number of the search targets is within the second threshold range, then the partition type of the original partition with the number of search targets within the second threshold range is determined as the target medium distribution area; When the number of the search targets is within the third threshold range, the partition type of the original partition with the number of search targets within the third threshold range is determined as the target dense distribution area; The determining, according to the partition type, the candidate frame search step size in each original partition includes: Setting the candidate frame search step size in the original partition where the partition type is the target sparse distribution area to the first step size; Setting the candidate frame search step size in the original partition where the partition type is the target medium distribution area to the second step size; Setting the candidate box search step size in the original partition where the partition type is the target dense distribution area to the third step size; The first threshold range is smaller than the second threshold range, the second threshold range is smaller than the third threshold range, the third step size is smaller than the second step size, and the second step size is smaller than the Describe the first step. 2. The method according to claim 1, wherein before the acquiring the image to be searched, further comprising: Get the original partition within the original image; collecting the occurrence frequency and distribution density of search targets in the original partition; According to the occurrence frequency and density of the search target, a Poisson distribution function of the original partition is determined. 3. A device for obtaining a candidate frame search step size, wherein the device comprises: a first acquisition unit, used to acquire the image to be searched; a second acquiring unit, configured to acquire the image information of each original partition in the image to be searched and the Poisson distribution function corresponding to each original partition; a first determining unit, configured to determine the partition type of each original partition according to the image information and Poisson distribution function of each original partition; a second determining unit, configured to determine, according to the partition type, a candidate frame search step size in each original partition; The first determining unit includes: The first determination module is used to determine the number of search targets in each original partition according to the image information and Poisson distribution function of each original partition; The second determination module is configured to determine the partition type of the original partition with the number of the search targets within the first threshold range as the target sparse distribution area when the number of the search targets is within the first threshold range; The third determination module is used to determine the partition type of the original partition with the number of the search targets within the second threshold range as the target medium distribution area when the number of the search targets is within the second threshold range; a fourth determining module, configured to determine the partition type of the original partition with the number of the search targets within the third threshold range as the target dense distribution area when the number of the search targets is within the third threshold range; the The second determination unit includes: The first setting module is used to set the candidate frame search step size in the original partition of the target sparse distribution area as the first step size; The second setting module is used to set the candidate frame search step size in the original partition where the partition type is the target medium distribution area to the second step size; The third setting module is used to set the candidate frame search step size in the original partition where the partition type is the target dense distribution area to the third step size; The first threshold range is smaller than the second threshold range, the second threshold range is smaller than the third threshold range, the third step size is smaller than the second step size, and the second step size is smaller than the Describe the first step. 4. The apparatus according to claim 3, wherein the apparatus further comprises: The third acquisition unit is used to acquire the original partition in the original image; a collection unit, configured to collect the occurrence frequency and distribution density of the search targets in the original partition; The third determining unit is configured to determine the Poisson distribution function of the original partition according to the occurrence frequency and density of the search target.

Images