JP2023536025A - Target detection method, device and roadside equipment in road-vehicle cooperation - Google Patents

Abstract

The present disclosure provides a target detection method, device, and roadside equipment in road-vehicle coordination, and relates to the field of intelligent transportation, and particularly to the field of image detection technology. A specific technical proposal is to perform target detection on an image, obtain a candidate target area in the image, a reliability value of the candidate target area, a degree of occlusion of the candidate target area, and detect the candidate target area. updating a confidence value of a candidate target region based on the IoU between regions and the degree of occlusion of the candidate target region; and extracting a target in the image from the candidate target region based on the updated confidence value. and detecting. When target detection is performed using the technical solution according to the embodiment of the present disclosure, the accuracy of target detection can be improved. [Selection diagram] Figure 1

Description

Cross-reference to related applications

This application is a Chinese patent application with application number 202110721853.4 and titled "Target Detection Method, Device and Roadside Device in Road-Vehicle Cooperation" filed with the Chinese Patent Office on June 28, 2021. , the entire contents of which are incorporated herein by reference.

The present disclosure relates to the field of intelligent traffic technology, and more particularly to the field of image detection technology.

In application scenarios such as road monitoring and vehicle route planning using road-to-vehicle cooperation V2X (Vehicle to everything, wireless communication technology for vehicles), after acquiring an image collected by an image collection device, people, animals, and vehicles in the image are identified. It is necessary to locate a target in an image by detecting such a target, and to trigger further processing operations on said target, or to link said targets to perform such things as vehicle path planning. Therefore, there is a need for a target detection method in road-vehicle cooperation for detecting targets in images.

The present disclosure provides a target detection method, apparatus and roadside equipment in road-vehicle cooperation.

One aspect of the present disclosure provides a target detection method in road-vehicle cooperation. The method includes:
performing target detection on an image to obtain a candidate target region in the image, a confidence value for the candidate target region, and an occlusion degree for the candidate target region;
updating the confidence values of the candidate target regions based on the IoU (Intersection over Union, evaluation index in object detection) between the candidate target regions and the degree of occlusion of the candidate target regions;
detecting targets in the image from candidate target regions based on the updated confidence values.

One aspect of the present disclosure provides a target detection apparatus in road-vehicle cooperation. The device comprises:
an information acquisition module for performing target detection on an image to acquire candidate target regions in the image, confidence values for the candidate target regions, and occlusion degrees for the candidate target regions;
a confidence value update module for updating the confidence values of the candidate target regions based on the IoU between the candidate target regions and the degree of occlusion of the candidate target regions;
a target detection module for detecting targets in the image from the candidate target regions based on the updated confidence values.

In another aspect of the disclosure, an electronic device is provided. The electronic device
at least one processor;
a memory in communication with the at least one processor;
The memory stores instructions to be executed by the at least one processor, and the instructions are executed by the at least one processor to instruct the at least one processor to perform a target detection method in road-vehicle cooperation. make it happen.

Another aspect of the present disclosure provides a non-transitory computer-readable storage medium having computer instructions stored thereon. The computer instructions are used to cause the computer to perform a target detection method in road-vehicle coordination.

According to another aspect of the disclosure, a computer program product is provided. The computer program product comprises a computer program, and when the computer program is executed by a processor, implements a target detection method in road-vehicle cooperation.

Another aspect of the present disclosure provides roadside equipment including the electronic device described above.

Another aspect of the present disclosure provides a crowd control platform including the electronic device described above.

As can be seen from the above, when performing target detection using the technical solution according to the embodiments of the present disclosure, first, based on the IoU between the candidate target regions and the degree of occlusion of the candidate target regions, the reliability of the candidate target regions and then detect targets in the image from the candidate target regions based on the updated confidence values. Since the IoU between candidate target areas can reflect the degree of overlap between each candidate target area, and the degree of occlusion of a candidate target area can reflect the degree of occlusion of the candidate target area, based on the IoU and the degree of occlusion: can refer to the degree of overlap between the target regions when updating the confidence values of the candidate target regions. By performing target detection on an image, the accuracy of target detection can be improved.

It will be understood that nothing described in this section is intended to identify key features or key features of embodiments of the disclosure, nor is it intended to limit the scope of the disclosure. Other features of the present disclosure will be readily understood by the following specification.

The following drawings are for better understanding of the present technical solution and are not limiting for the present disclosure.
FIG. 1 is a schematic flow chart of a target detection method in road-vehicle cooperation according to an embodiment of the present disclosure. FIG. 2 is a schematic diagram of an image according to an embodiment of the present disclosure; FIG. 3a is a schematic diagram of the structure of a network model according to an embodiment of the present disclosure; FIG. 3b is a schematic diagram of another network model structure according to an embodiment of the present disclosure. FIG. 4 is a schematic diagram of the structure of a target detection device in road-vehicle cooperation according to an embodiment of the present disclosure. FIG. 5 is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure.

Illustrative embodiments of the present disclosure will now be described with reference to the drawings. Various details of embodiments of the disclosure are included for ease of understanding of the disclosure and should be considered as exemplary only. As such, those skilled in the art will appreciate that various changes and modifications can be made to the examples described herein without departing from the scope and spirit of this disclosure. Also, in the following description, descriptions of well-known functions and structures are omitted for clarity and brevity.

Embodiments of the present disclosure provide a target detection method, apparatus and roadside equipment in road-vehicle coordination.

An embodiment of the present disclosure provides a target detection method for road-vehicle cooperation. The method is
performing target detection on the image, a candidate target region in the image, a confidence value for the candidate target region;
obtaining the degree of occlusion of the candidate target area;
updating the confidence values of the candidate target regions based on the IoU (Intersection over Union, also called overlap ratio) between the candidate target regions and the degree of occlusion of the candidate target regions;
detecting a target in the image from the candidate target regions based on the updated confidence values;
including.

The IoU between candidate target areas can reflect the degree of overlap between each candidate target area, and the degree of occlusion of a candidate target area can reflect the degree of occlusion of the candidate target area. When updating the confidence values of the candidate target regions by using the Target detection can be performed on the image by the value to increase the accuracy of the target detection.

An execution subject according to an embodiment of the present disclosure will be described below.

An execution subject according to an embodiment of the present disclosure may be an electronic device equipped with a target detection function. Here, the electronic device may be a desktop computer, a notebook computer, a server, an image acquisition device, or the like. Among them, the image collecting device may include a video camera, a camera, a driving recorder, and the like.

The technical solutions according to the embodiments of the present disclosure are used to perform target detection on collected images in application scenarios such as road monitoring by road-vehicle cooperation V2X, vehicle route planning, and so on.

The technical solution according to the embodiments of the present disclosure can also be used to perform target detection on the collected images in other scenarios. For example, the other scenarios mentioned above may be crowded scenarios such as subway stations, department stores, concerts, and the like. When images are collected for such a scenario, the people included in the collected images are often densely packed, and it is likely that some people's faces will be blocked by other people's faces. Also, the above scenarios may be scenarios where people are relatively dense, such as museum entrances and bank lobbies. When image acquisition is performed for such a scenario, a person's face may be obscured by other people, buildings, or the like in the acquired image.

The above is merely an example of application scenarios according to embodiments of the present disclosure and is not intended to limit the present disclosure.

The target may be a human face, an animal, a vehicle, or the like.

Hereinafter, a target detection method in road-vehicle cooperation according to an embodiment of the present disclosure will be specifically described.

Referring to FIG. 1, FIG. 1 is a schematic flow chart of a road-vehicle cooperation target detection method according to an embodiment of the present disclosure, which includes steps S101 to S103 as follows.

In step S101, target detection is performed on the image to obtain candidate target regions in the image, reliability values of the candidate target regions, and occlusion degrees of the candidate target regions.

The images may be images obtained by image acquisition for a specific scenario. The scenarios may include vehicle driving scenarios, parking lot scenarios, etc. In this case, the target may be a vehicle. The scenarios may also include scenarios of public spaces such as subway stations or high-speed train stations, in which case the target may be a person.

When target detection is performed, in one embodiment target detection is performed on the image using a preset target detection algorithm to determine candidate target regions in the image, confidence values for the candidate target regions, candidate target regions may be acquired.

The preset target detection algorithms may be detection algorithms employed for different types of targets. For example, if the target is a person, a face detection algorithm, a human body detection algorithm, or the like may be employed. If the target is a vehicle, vehicle detection algorithms, license plate detection algorithms, etc. may be employed.

Other embodiments of target detection can be found in the examples below and will not be described in detail here.

A candidate target region refers to a region where target detection determines that a target may be present. Taking FIG. 2 as an example, the areas surrounded by respective rectangular frames in FIG. 2 are candidate target areas obtained by performing animal detection on the image.

A confidence value for a candidate target area reflects the likelihood that a target exists in the candidate target area. The confidence value may be expressed as a decimal number, percentage, or the like. A higher value of the confidence value indicates a higher probability that a target is present in the candidate target region.

For example, if the target is a person, the confidence value for candidate target area A being greater than the confidence value for candidate target area B indicates the likelihood that a person is present in candidate target area A. Indicates more than likely to exist.

The degree of occlusion of a candidate target area reflects the extent to which the candidate target area is occluded. The degree of shielding may be represented by a decimal number, a percentage, or the like, or may be represented by a shielding level number. For example, if the shielding level numbers include 1, 2, and 3, number 1 indicates that the shielding level is heavy shielding, number 2 indicates that the shielding level is medium shielding, and number 3 indicates that the shielding level is medium shielding. may represent that the shielding level is light shielding.

In step S102, the reliability values of the candidate target regions are updated based on the IoU between the candidate target regions and the degree of occlusion of the candidate target regions.

The IoU between candidate target areas is to represent the degree of overlap between two candidate target areas. A higher IoU indicates a higher degree of overlap between the two candidate target regions. A lower IoU indicates less overlap between the two candidate target regions.

Specifically, the overlapping area between two candidate target regions is calculated to obtain a first area, the sum of the areas of the two candidate target regions is calculated to obtain a second area, and the second area is and the first area to obtain a third area, and determine the ratio of the first area to the third area as the IoU between the candidate target regions.

For example, if the area of candidate target area A is 48 and the area of candidate target area B is 32, then the overlapping area of candidate target area A and candidate target area B is 16, i.e. the first area is 16. be. The sum of the area of candidate target area A and the area of candidate target area B is (46+32)=80, ie the second area is 80; The result of calculating the difference between the second area and the first area is (80-16)=64, ie the third area is 64. The result of calculating the ratio of the first area to the third area is 16/64=0.25, where 0.25 is the IoU between the candidate target regions.

In one embodiment, a reference region is selected from each candidate target region, and for each other candidate target region other than the reference region in each candidate target region, the I
The oU may be calculated and the calculated IoU may be determined as the IoU for updating the confidence value for that candidate target region.

The reference region may be the region with the highest confidence value in each candidate target region.

In another embodiment, for each candidate target area, select one IoU from the IoUs between that candidate target area and each of the other candidate target areas, and assign the selected IoU to the confidence of that candidate target area. It can also be determined as an IoU for updating the degree value.

For example, a maximum IoU, an average IoU, a median IoU, a minimum IoU, etc. may be selected from the plurality of IoUs.

When updating the confidence values of the candidate target regions, in one embodiment, they are adjusted according to the first and second preset weights based on the IoU between the candidate target regions and the degree of occlusion of the candidate target regions. A factor may be calculated and the confidence value for the candidate target region may be updated based on the calculated adjustment factor.

Specifically, calculating the product between the IoU between the candidate target areas and the first weight, calculating the product between the degree of occlusion of the candidate target area and the second weight, and calculating the product of the two products obtained by the calculation. Let the sum be the adjustment factor.

For example, the IoU between the candidate target areas is 80%, the degree of occlusion of the candidate target areas is 50%, the preset first weight is 0.8, and the preset second weight is 0. .2, the result of computing the product between the IoU between the candidate target regions and the first weight is 0.8*80%=6
4%, and the result of calculating the product between the degree of occlusion of the candidate target area and the second weight is 0.2*50
Since %=10% and the sum of the two products obtained by calculation is 64%+10%=74%, we obtain an adjustment factor of 74%.

After the adjustment factor is obtained from the calculation, the product between the adjustment factor and the confidence value of the candidate target region may be calculated to be the updated confidence value of the candidate target region.

Other embodiments for updating the confidence values of candidate target regions can be found in the examples below and will not be described in detail here.

In step S103, targets in the image are detected from the candidate target regions based on the updated confidence values.

In one embodiment of the present disclosure, candidate target regions with updated confidence values greater than a preset confidence threshold may be selected, and targets in the selected candidate target regions may be established as targets in the image.

The preset reliability threshold is set by the operator based on experience. For example, if the confidence value is expressed as a percentage, the preset confidence threshold may be 90%, 95%, and so on.

An example will be given to explain the process of determining the above target. Suppose the confidence value of each updated candidate target region is 80%, 70%, 90%, 95%, respectively, the preset confidence threshold is 85%, and the updated value is greater than 85%. 90%, 95%, where the updated confidence value for region 1 is 90% and the updated confidence value for region 2 is 95%, then region 1 The target in and the target in region 2 are the targets in the image.

Thus, for candidate target regions whose confidence values are greater than the preset confidence threshold, the probability that these candidate target regions contain the target is higher than the probability that other candidate target regions contain the target. . Therefore, targets in candidate target regions with confidence values greater than a preset confidence threshold are determined as targets in the image, and the accuracy of the obtained targets is increased.

In one embodiment of the present disclosure, a preset number of candidate target regions with the highest updated confidence values may be selected, and targets in the selected candidate target regions may be established as targets in the image.

The preset number may be empirically set by the operator. For example, the preset number may be one, three, five, or the like.

An example will be given to explain the process of determining the above target. Assuming that the reliability values of the target regions are 80%, 70%, 90%, and 95%, respectively, and the preset number is 3, the top 3 with the highest updated reliability values are , 95%, 90%, and 80%, respectively, and establish targets in candidate target regions with updated confidence values of 95%, 90%, and 80% as targets in the image.

Thus, for a preset number of candidate target areas starting with the highest confidence value, the probability that these candidate target areas contain the target is higher than the probability that other candidate areas contain the target. Thus, targets in a preset number of candidate target regions starting with the highest confidence value are established as targets in the image, and the accuracy of the obtained targets is increased.

As can be seen from the above, when performing target detection using the technical solution according to the embodiments of the present disclosure, first, based on the IoU between the candidate target regions and the degree of occlusion of the candidate target regions, the reliability of the candidate target regions and then detect targets in the image from the candidate target regions based on the updated confidence values. Since the IoU between candidate target areas can reflect the degree of overlap between each candidate target area, and the degree of occlusion of a candidate target area can reflect the degree of occlusion of the candidate target area, based on the IoU and the degree of occlusion: can refer to the degree of overlap between the candidate target regions when updating the confidence values of the candidate target regions. Target detection can be performed on the image based on the degree value to increase the accuracy of target detection.

It should be noted that the target occluded situation is particularly serious in high-density scenarios such as, for example, high-density scenarios of people and high-density vehicles. For images in these dense scenarios, the occlusion of each candidate target region is high, so the targets within the candidate target regions are imperfect and the confidence values obtained for the candidate target regions have large errors. Updating the confidence values of the candidate target regions by the degree of occlusion of the candidate target regions effectively removes the effect of errors on the confidence values when each candidate target region is occluded, thus the updated confidence The accuracy of the degree value can be increased and the detection can also obtain a precise target. Therefore, the technical solutions according to the embodiments of the present disclosure can be better applied in the case of occlusion in dense scenarios, and can enhance the accuracy of target detection.

To accurately update the confidence values of candidate target regions, one embodiment of the present disclosure selects the first region with the highest confidence value from the region set, and compares the first region with the other regions in the region set. update the confidence values of other regions according to the IoU between and the degree of occlusion of other regions, and complete one confidence value update operation. The above operation is repeated until the region set contains one region. One reliability value update operation may be referred to as one cycle.

The region set includes non-selected regions within the candidate target regions. Specifically, at the beginning of the first cycle, the region set includes each candidate target region obtained in step S101. After selecting the first region from the region set in each cycle, the previously selected first region is no longer included in the region set.

At the beginning of the first cycle, the first region is the region with the highest confidence value for each candidate target region obtained in step S101. In each subsequent cycle, the first region is the region with the highest confidence value in each updated region obtained by the previous cycle.

The other area is an area other than the first area in the area set. For example, the region set includes region 1, region 2, region 3, where region 1 is the first region and region 2, region 3 are regions other than the first region, then region 2, region 3 is the other area.

In each cycle, over each region in the region set, the confidence value of each region may be sorted from highest to lowest, and the region with the highest confidence value may be determined as the first region. Also, the first regions may be stored in the prediction set, and as the number of cycles increases, the number of first regions stored in the prediction set also increases.

The process of the above cycle will now be described with reference to a specific example.

Suppose each candidate target area obtained in step S101 is b1, b2, b3, . . . bn.

At the beginning of the first cycle, region set B={b1, b2, b3, . . . , bn}. Here, since the region b1 has the highest reliability value among the candidate target regions, the region b1 is defined as the first region. Since the areas other than the area b1 in the area set B are b2, b3, . . . , bn, {b2, b3, .

The IoU between the other areas {b2, b3, . . . , bn} and the area b1, and the other areas {b2, b3
, bn}, the reliability values of the other regions {b2, b3, . . . , bn} are updated. Then, the first region b1 may be added to the prediction set D, and the added prediction set D={b1}.

At the start of the second cycle, region b1 has already been selected as the first region, so region set B does not include region b1, and region set B={b2, b3, . . . , bn}. Here, since the region b2 has the highest reliability value in the updated {b2, b3, . . . , bn}, the region b2 is defined as the first region. Since the areas other than the area b2 in the area set B are b3, . . . , bn, {b3, .

Based on the IoU between the other regions {b3, . Update the confidence value. Then, the first region b2 may be added to the prediction set D, and the added prediction set D={b1, b2}.

At the start of the third cycle, the region set B={b3, . where the updated {b3,...,bn}
Since the region b3 has the highest reliability value in , the region b3 is set as the first region. Since the regions other than region b3 in region set B are b4, . . . , bn, {b4, .
is the other area.

Based on the IoU between the other regions {b4, . . . , bn} and the region b3 and the degree of shielding of the other regions {b4, . Update the confidence value. Then, the first region b3 may be added to the prediction set D, and the added prediction set D={b1, b2, b
3}.

In a similar fashion, the above process is repeated until the number of regions in region set B is 1, adding only one region in region set B directly to prediction set D, the cycle ends and each updated region Confidence value obtained.

Thus, in each cycle, the confidence values of the regions in the region set are updated based on the IoU between the other regions in the region set and the first region and the degree of occlusion of the other regions. Here, the degree of occlusion of other regions reflects the degree of occlusion of other regions, and when a region is occluded, the accuracy of the reliability value of the region obtained by detection is low. The accuracy of the updated candidate target region confidence values can be increased by introducing the occlusion of the regions of . Since the IoU between the other region and the first region reflects the degree of overlap between the other region and the first region, and the first region is the region with the highest reliability value, , the reliability values of other regions can be effectively adjusted according to the degree of overlap with the region with the highest reliability value. Therefore, in each cycle, the reliability values of other regions can be effectively updated based on the IoU and the degree of shielding. By repeating the updating process, the accuracy of the updated reliability value can be further improved.

In one embodiment of the present disclosure, updating the reliability values of other regions in each cycle can be achieved by the following steps A1-A4.

In step A1, the IoU between the other regions in the region set and the first region is calculated.
Specifically, first, the overlapping area between the other area and the first area is calculated, and the total area of the other area and the first area is calculated. Next, the difference between the sum of the areas and the overlapping area is calculated to obtain the target area, and the ratio between the overlapping area and the target area is established as the IoU between the candidate target regions.

At step A2, a first reliability adjustment value is determined based on the IoU and a preset IoU threshold.
The preset IoU threshold can be set by the operator based on experience. For example, the IoU threshold may be 90%, 95%, and so on.

In one embodiment, determining whether the IoU is less than a preset IoU threshold, if the IoU is less than the preset IoU threshold, determine a first reliability value adjustment value as a first preset value; If the IoU is greater than or equal to a preset IoU threshold, the first reliability adjustment value may be determined as the second preset value.
Both the first preset value and the second preset value are set by the operator based on experience.

In an embodiment of the present disclosure, it is determined whether the IoU is less than a preset IoU threshold, and if the IoU is less than the preset IoU threshold, the first reliability adjustment value is determined as 1, and the IoU is greater than or equal to a preset IoU threshold, a first confidence adjustment value may be established as the difference between 1 and the IoU.
For example, if the preset IoU threshold is 90%, and the IoU between the other area and the first area is 95%, IoU 95% is the preset IoU threshold 90%.
is greater than, determine that the first reliability adjustment value is 1-95%=5%, and if the IoU between the other region and the first region is 50%, IoU 50% is preset Since it is less than the IoU threshold of 90%, the first reliability adjustment value is determined as 1.

Thus, if the IoU is less than the preset IoU threshold, it indicates that the degree of overlap between the other region and the first region is small, and a small portion of the image content in the other region is masked and detected. means that the reliability value of the other region obtained by is highly accurate, in which case the reliability value of the other region need not be adjusted. By setting the first reliability adjustment value to 1, it is possible to realize that the reliability value of the region is not adjusted. If the IoU is greater than or equal to the preset IoU threshold, it indicates that the degree of overlap between the other region and the first region is large, and most of the image content in the other region is occluded, resulting in the detection It means that the accuracy of the reliability value of the other region is low, in which case the reliability value of the other region needs to be adjusted, and the first reliability adjustment value is set as the difference between 1 and IoU This allows the adjusted confidence values to be closer to the actual situation.

In step A3, a second reliability adjustment value is determined based on the degree of shielding of other regions.

In one embodiment, the second reliability adjustment value is calculated by multiplying the degree of occlusion of the other regions and a preset adjustment factor.

The preset adjustment coefficient may be set by the operator based on experience, and for example, the preset adjustment coefficient may be 1.2, 1.5, or the like.

In one embodiment of the present disclosure, the second reliability adjustment value g(occ_pred) can also be determined according to the following equation.
Here, occ_pred is the degree of occlusion of other regions, α is a preset constant, and α>1.

Since α>1, the second reliability adjustment value g(occ_pred) increases as the degree of occlusion of other regions increases.

If the occlusion of an area is high, the accuracy of the confidence value of the area is low, so the confidence value of the area can be adjusted significantly to bring the adjusted confidence value closer to the actual situation. is necessary. In addition, the second reliability adjustment value g(occ_pred) increases as the degree of occlusion of other regions increases. By adjusting the confidence values of the other regions by a large amount, the adjusted confidence values of the other regions can be closer to the actual situation.

In step A4, the reliability values of other regions are adjusted using the first reliability adjustment value and the second reliability adjustment value.

In one embodiment of the present disclosure, confidence values for other regions may be adjusted according to the following formula.
S'=S*T1*T2
Here, S′ represents the confidence value of the adjusted other region, S represents the confidence value of the other region before adjustment, T1 represents the first confidence adjustment value, and T2 represents the second confidence value. 2 Represents the confidence adjustment value.

Thus, since the adjusted confidence value is the product of the first confidence adjustment value, the second confidence adjustment value, and the confidence value of the other region, the first confidence value Since the adjustment value and the second reliability adjustment value reflect the occlusion situation of other areas from different angles, the adjusted reliability value can refer to the occlusion situation of other areas. , to bring the adjusted confidence values closer to the real situation.

In another embodiment, the reference reliability value is calculated by multiplying the first reliability adjustment value, the second reliability adjustment value, and the reliability value of the other region to obtain the preset reliability value. The reference reliability value may be adjusted according to the degree error value to obtain an adjusted reference reliability value, which is used as an adjusted reliability value for another region.
Specifically, the product between the preset reliability value error value and the parameter reliability value can be calculated, and the product obtained by the calculation can be determined as the adjusted reliability value of the other region.

Thus, since the first reliability adjustment value is determined by the IoU between the other region and the first region, and the IoU reflects the degree of overlap between the other region and the first region, and 2 Since the reliability adjustment value is determined by the degree of occlusion of other regions, and the degree of occlusion reflects the degree of occlusion of other regions, the first reliability adjustment value and the second reliability adjustment value are Both can reflect the occluded situation of other regions from different angles. Therefore, when adjusting the reliability values of other regions using the first reliability adjustment value and the second reliability adjustment value, the first reliability adjustment value and the second reliability adjustment value are obtained from different angles. Since it reflects the occlusion situation of other areas, if the first reliability adjustment value and the second reliability adjustment value are used for adjustment, the reliability can be more accurately determined based on the occlusion situation of other areas. Adjusting the confidence value brings the adjusted confidence value closer to the actual situation.

Hereinafter, a method for updating the reliability value in the above cycle will be described according to a specific implementation process.

Suppose each candidate target area is b1, b2, b3, the preset IoU threshold Nt is 90%, and the reliability value and occlusion degree of each candidate target area are shown in Table 1 below.

At the beginning of the first cycle, region set B={b1, b2, b3}, where region b1 has the highest confidence value Cv1, region b1 is the first region, regions b2, b3 are Another area.

For the area b2, calculate the IoU between the area b2 and the area b1, and determine the first reliability adjustment value based on the IoU and the preset IoU threshold of 90%. A second reliability adjustment value is determined based on the degree of shielding Co2 of the region b2. The reliability value Cv2 of the region b2 is adjusted based on the first reliability adjustment value and the second reliability adjustment value, and the updated reliability value is Cv21.

For the area b3, calculate the IoU between the area b3 and the area b1, and determine the first reliability adjustment value based on the IoU and the preset IoU threshold of 90%. A second reliability adjustment value is determined based on the shielding degree Co3 of the region b3. The reliability value Cv3 of the region b3 is adjusted based on the first reliability adjustment value and the second reliability adjustment value, and the updated reliability value is Cv31.

The updated confidence values for each candidate target region obtained in the first cycle are shown in Table 2 below.

At the beginning of the second cycle, region b1 was already selected, so region set B = {b2, b
3}. Here, the region b2 has the highest reliability value Cv21, the region b2 is the first region, and the region b3 is the other region.

For the area b3, calculate the IoU between the area b3 and the area b2, and determine the first reliability adjustment value based on the IoU and the preset IoU threshold of 90%. A second reliability adjustment value is determined based on the shielding degree Co3 of the region b3. The reliability value Cv31 of the region b3 is adjusted based on the first reliability adjustment value and the second reliability adjustment value, and the updated reliability value is Cv311.

Since regions b1, b3 have already been selected, region set B={b3}, containing one region, the cycle ends.

The resulting updated confidence value for each candidate target region is shown in Table 3 below.

In one embodiment of the present disclosure, for each different target scale in step S101 above, target detection is performed on the image, and the candidate target regions of different scales in the image, the confidence values of the candidate target regions, and the candidate target regions can be obtained.

Target scale refers to the dimensions of a target.
The target scale may be a preset scale value, for example, the target scale may be 16x16, 32x32, 64x64.

Specifically, multi-layer feature extraction may be performed on the image, and feature fusion may be performed on different features to obtain features of different scales. Target detection is performed on the image using features of different scales to obtain candidate target regions of different scales, and confidence values and degrees of occlusion of the candidate target regions of different scales are obtained.

Thus, because candidate target regions of different scales contain different image feature information, obtaining candidate target regions of different scales in the image enhanced the feature information of the candidate target regions at different scales.

In one embodiment of the present disclosure, an image is input to a pre-trained target detection model, and candidate target regions in the image output by the target detection model, confidence values for the candidate target regions, and candidate targets The degree of occlusion of the area can be obtained.

The target detection model includes a target detection layer for detecting candidate target regions in an image and an occlusion prediction layer for predicting occlusion of candidate target regions.

In one embodiment, the target detection layer, in addition to detecting candidate target regions within an image, may also compute confidence values for the candidate target regions. In this case, the network structure of the target detection model is shown as in Fig. 3a. The target detection model includes a target detection layer and a shielding degree prediction layer.

Specifically, after an image is input to the target detection model, the target detection layer in the model detects candidate target regions in the image, calculates confidence values for the candidate target regions, and sends the detection results to the occlusion prediction layer. transmit to The occlusion prediction layer predicts the occlusion of each candidate target region. The target detection model outputs candidate target regions and confidence values and degrees of occlusion for the candidate target regions.

As can be seen from the above-described embodiments, when performing target detection on an image, for each target scale, candidate target regions with different scales, reliability values of the candidate target regions, and degrees of occlusion of the candidate target regions are used. to get In this case, FPN (Feature Pyr
amid Networks), and the FPN is used to obtain various different scales of candidate target regions, as well as confidence values and occlusions of candidate target regions.

The network structure of the network model with the addition of FPN is shown in Fig. 3b. The network structure shown in FIG. 3b includes a Backbone and an FPN.
Here, the backbone is used to perform feature extraction on the image, obtain the image features of different layers in the image, and input the image features of different layers to the FPN.

For example, if the backbone is a convolutional neural network, each convolutional layer of the convolutional neural network can perform a convolution operation on an image to obtain image features of different layers.

FPN performs feature fusion on image features of different layers, obtains image features of different scales, performs target detection based on the image features of different scales, obtains candidate target regions of different scales, obtains candidate targets It is used to obtain the confidence value and occlusion of a region, and implements the divide-and-conquer process for image features in different layers.

During target detection, the sample image is taken as the training sample, and the true candidate target region and the true degree of occlusion in the sample image are taken as training labels for the preset neural network model until the training termination condition is met. Train and get a trained target detection model.

The preset neural network model is CNN (Conv Neura
l Network) model, RNN (Recurrent Neural Network)
k) It may be a model, a DNN (Deep Neural Network) model, or the like.

Specifically, after a sample image is input to a preset neural network model, the preset neural network model performs target detection on the sample image to determine candidate target regions and occluded regions in the sample image. and calculating the difference between the candidate target area and the true target area and the difference between the occlusion of the candidate target area and the true occlusion, and based on the calculated difference to adjust the parameters of the neural network model, and repeat parameter adjustment until a preset training termination condition is satisfied.

The training termination condition may be that the number of times of training reaches a preset number, that the model parameters satisfy a preset convergence condition of the model parameters, or the like.

Since the target detection model was trained with a large amount of training samples, in the training process, the target detection model learned features of target regions and occluded features in sample images. Therefore, since the target detection model has high robustness, when target detection is performed on an image by the target detection model, it outputs accurate candidate target regions, reliability values, and occlusion degrees of the candidate target regions. can be done.

In step S101, in addition to performing target detection on the image using the target detection model, the image is segmented into a plurality of regions, image features in each region are extracted, and region detection is performed based on the image features. Determine candidate target regions from .

The image features include texture features, color features, edge features, and the like.

After obtaining each candidate target area, predict a confidence value for each candidate target based on the image characteristics of each candidate target area.

Then, the occlusion degree of each candidate target area can be calculated based on the layer to which each candidate target area belongs and the position information.

Specifically, based on the relative relationship between the layer to which the candidate target area belongs and the position, it is determined whether there is a shield between the candidate target areas, and the area between the shielded area and the area of the shielded area is determined. is calculated as the occlusion degree of the candidate target area.

For example, if candidate target area A is on the foreground layer, candidate target area B is on the background layer, and there is an overlap between the location information of candidate target area A and the location information of candidate target area B, then candidate target area B is It can be determined that it is occluded, and the ratio of the occluded area of the candidate target area B to the area of the candidate target area B is calculated as the occluded degree of the candidate target area B;

Corresponding to the above target detection method in road-vehicle cooperation, the embodiments of the present disclosure further provide a target detection device in road-vehicle cooperation.

Referring to FIG. 4, FIG. 4 is a structural schematic diagram of a target detection device in road-vehicle cooperation according to an embodiment of the present disclosure, the device includes the following modules 401-403.
The information acquisition module 401 is for performing target detection on an image to acquire candidate target regions in said image, confidence values of the candidate target regions, and occlusion degrees of the candidate target regions.
The confidence value update module 402 is for updating the confidence values of the candidate target regions based on the IoU between the candidate target regions and the degree of occlusion of the candidate target regions.
A target detection module 403 is for detecting a target in said image from the candidate target regions based on the updated confidence value.

As can be seen from the above, when performing target detection using the technical solution according to the embodiments of the present disclosure, first, based on the IoU between the candidate target regions and the degree of occlusion of the candidate target regions, the reliability of the candidate target regions and then detect targets in the image from the candidate target regions based on the updated confidence values. Since the IoU between candidate target areas can reflect the degree of overlap between each candidate target area, and the degree of occlusion of a candidate target area can reflect the degree of occlusion of the candidate target area, based on the IoU and the degree of occlusion: Therefore, the overlapping degree between the target regions can be referred to when updating the confidence value of the candidate target region, so that the updated confidence value of the candidate target region is more closely related to the actual situation. Therefore, target detection can be performed on the image based on the updated reliability value, and the accuracy of target detection can be improved.

In one embodiment of the present disclosure, the confidence value update module 402 specifically selects a first region with the highest confidence value from a region set, and compares other regions in the region set with the first region. It is used to repeat updating the confidence values of other regions based on the IoU between them and the degree of occlusion of other regions until a region is included in the region set. Here, the region set includes non-selected regions in the candidate target regions.

Thus, in each cycle, the confidence values of the regions in the region set are updated based on the IoU between the other regions in the region set and the first region and the degree of occlusion of the other regions. Here, the degree of occlusion of other regions reflects the degree of occlusion of other regions, and when a region is occluded, the accuracy of the reliability value of the region obtained by detection is low. The accuracy of the updated candidate target region confidence values can be increased by introducing the occlusion of the regions of . Since the IoU between the other region and the first region reflects the degree of overlap between the other region and the first region, and the first region has the highest reliability value, , the reliability values of other regions can be effectively adjusted according to the degree of overlap with the region with the highest reliability value. Therefore, in each cycle, the reliability values of other regions can be effectively updated based on the IoU and the degree of shielding. By repeating the updating process, the accuracy of the updated reliability value can be further improved.

In one embodiment of the present disclosure, the confidence value update module 402 includes:
IoU calculation means for calculating the IoU between other regions in the region set and the first region;
a first adjustment value determination means for determining a first reliability adjustment value based on the IoU and a preset IoU threshold;
a second adjustment value determining means for determining a second reliability adjustment value based on the degree of shielding of other regions;
confidence value adjustment means for adjusting confidence values of other regions using the first confidence adjustment value and the second confidence adjustment value.

Thus, since the first reliability adjustment value is determined by the IoU between the other region and the first region, and the IoU reflects the degree of overlap between the other region and the first region, and 2 Since the reliability adjustment value is determined by the degree of occlusion of other regions, and the degree of occlusion reflects the degree of occlusion of other regions, the first reliability adjustment value and the second reliability adjustment value are Both can reflect the occluded situation of other regions from different angles. Therefore, when adjusting the reliability values of other regions using the first reliability adjustment value and the second reliability adjustment value, the first reliability adjustment value and the second reliability adjustment value are obtained from different angles. Since it reflects the occlusion situation of other areas, if the first reliability adjustment value and the second reliability adjustment value are used for adjustment, the reliability can be more accurately determined based on the occlusion situation of other areas. Adjusting the confidence value brings the adjusted confidence value closer to the actual situation.

In an embodiment of the present disclosure, the first adjustment value determination means specifically determines whether the IoU is smaller than a preset IoU threshold, if less, determining a first reliability adjustment value as 1; and if the IoU is greater than or equal to a preset IoU threshold, determining the first reliability adjustment value as the difference between 1 and the IoU. , is used for

Thus, if the IoU is less than the preset IoU threshold, it indicates that the degree of overlap between the other region and the first region is small, and a small portion of the image content in the other region is masked and detected. means that the reliability value of the other region obtained by is highly accurate, in which case the reliability value of the other region need not be adjusted. By setting the first reliability adjustment value to 1, it is possible to realize that the reliability value of the region is not adjusted. If the IoU is greater than or equal to the preset IoU threshold, it indicates that the degree of overlap between the other region and the first region is large, and most of the image content in the other region is occluded, resulting in the detection It means that the accuracy of the reliability value of the other region is low, in which case the reliability value of the other region needs to be adjusted, and the first reliability adjustment value is set as the difference between 1 and IoU This allows the adjusted confidence values to be closer to the actual situation.

In one embodiment of the present disclosure, the second adjustment value determining means is specifically used to determine the second reliability adjustment value g(occ_pred) according to the following formula.
Here, occ_pred is the degree of occlusion of other regions, α is a preset constant, and α>1.

If the occlusion of an area is high, the accuracy of the confidence value of the area is low, so the confidence value of the area can be adjusted significantly to bring the adjusted confidence value closer to the actual situation. is necessary. In addition, the second reliability adjustment value g(occ_pred) increases as the degree of occlusion of other regions increases. By adjusting the confidence values of the other regions by a large amount, the adjusted confidence values of the other regions can be closer to the actual situation.

In one embodiment of the present disclosure, the reliability value adjusting means is specifically used to adjust the reliability value of other regions according to the following formula.
S'=S*T1*T2
Here, S′ represents the adjusted confidence value of the other region, S represents the confidence value of the other region before adjustment, T1 represents the first confidence adjustment value, and T2 is represents the second reliability adjustment value;

Thus, since the adjusted confidence value is the product of the first confidence adjustment value, the second confidence adjustment value, and the confidence value of the other region, also the first confidence adjustment Since the value and the second reliability adjustment value reflect the situation where other areas are blocked from different angles, the adjusted reliability value refers to the situation where other areas are blocked, Make the adjusted confidence values more closely match the actual situation.

In one embodiment of the present disclosure, the target detection module 403 specifically selects candidate target regions whose updated confidence values are greater than a preset confidence threshold, and detects targets in the selected candidate target regions. as a target in said image, or selecting a preset number of candidate target regions with the highest updated confidence value and establishing a target in the selected candidate target region as a target in said image. used for

Thus, for candidate target areas whose confidence values are greater than a preset confidence threshold, those candidate target areas are more likely to contain the target than other candidate target areas are likely to contain the target. Thus, when a target in a candidate target region with a confidence value greater than a preset confidence threshold is determined as a target in the image, the resulting target accuracy is high. For a preset number of candidate target areas starting with the highest confidence value, these candidate target areas are more likely to contain the target than other candidate areas to contain the target. Thus, if the targets in a preset number of candidate target regions starting with the highest confidence value are established as targets in the image, the accuracy of the targets obtained is high.

In one embodiment of the present disclosure, the information acquisition module 401 specifically performs target detection on an image for each different target scale, and includes candidate target regions of different scales in the image and target regions of the candidate target regions. It is used to obtain the confidence value and the degree of occlusion of the candidate target area.

Thus, because candidate target regions of different scales contain different image feature information, obtaining candidate target regions of different scales in the image enhances the feature information of the candidate target regions at different scales.

In one embodiment of the present disclosure, the information acquisition module 401 specifically inputs an image to a pre-trained target detection model, and calculates candidate target regions in the image output by the target detection model. is used to obtain the confidence value of the candidate target area and the occlusion degree of the candidate target area. Here, the target detection model includes a target detection layer for detecting candidate target regions in an image and a degree of occlusion prediction layer for predicting the degree of occlusion of the candidate target region.

Thus, during the training process, the target detection model learned features of target regions and occluded features in the sample images because the target detection model was trained with a large number of training samples. Therefore, since the target detection model has a high robustness, when the target detection model performs target detection on an image, it can output accurate candidate target regions, confidence values, and occlusion degrees of the candidate target regions. can.

In the technical solution of this disclosure, the collection, storage, use, processing, transmission, provision and disclosure of the user's personal information shall comply with relevant laws and regulations and shall not violate public order and morals. .

According to embodiments of the disclosure, the disclosure further provides devices, readable storage media, and computer program products.

SUMMARY In one embodiment of the present disclosure, an electronic device is provided.
The electronic device
at least one processor;
a memory in communication with the at least one processor;
The memory stores instructions to be executed by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform any of the foregoing method embodiments. or one target detection method in road-vehicle cooperation can be executed.

One embodiment of the present disclosure provides a non-transitory computer-readable storage medium having computer instructions stored thereon. The computer instructions are used to cause the computer to perform the target detection method in road-vehicle coordination of any one of the method embodiments described above.

In one embodiment of the disclosure, a computer program product is provided. The computer program product comprises a computer program, and when the computer program is executed by a processor, implements the target detection method in road-vehicle cooperation of any one of the aforementioned method embodiments.

An embodiment of the present disclosure provides roadside equipment including the above electronic device.

An embodiment of the present disclosure provides a cloud control platform including the above electronic device.

FIG. 5 shows a schematic block diagram of an exemplary electronic device 500 for implementing embodiments of the present disclosure. Electronic devices include various forms of digital computers such as, for example, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices also include various forms of mobile devices such as, for example, personal digital assistants, cell phones, smart phones, wearable devices, and other similar computing devices. The components, their connections, and their functionality described herein are exemplary only, and are not limiting of what is described and/or claimed herein with respect to the practice of this disclosure.

As shown in FIG. 5, device 500 can be operated in any suitable manner by a computer program stored in read only memory (ROM) 502 or loaded from storage means 508 into random access memory (RAM) 503 . It includes computing means 501 for performing operations and processes. The RAM 503 also contains the device 50
Various programs and data for manipulating 0 can also be stored. Computing means 501 , ROM 502 and RAM 503 are each connected by a bus 504 . Input/output (I/O) interface 505 is also connected to bus 504 .

Multiple components in device 500 are connected to I/O interface 505 . The plurality of components include input means 506 such as a keyboard and mouse, output means 507 such as various types of displays and speakers, storage means 508 such as disks and optical disks, and communication means 509 such as network cards, modems, and communication means 509, such as a wireless communication transceiver. Communication means 509 allows device 500 to exchange information/data with other devices, eg, via computer networks, such as the Internet, and/or various communication networks.

Computing means 501 may be various general purpose and/or special purpose processing components having processing and computing capabilities. Some examples of computing means 501 are a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signals Including, but not limited to, a processor (DSP), and any suitable processor, controller, microcontroller, or the like. The calculation means 501 executes each of the methods and processes described above, for example, the road-vehicle cooperation target detection method. For example, in some embodiments, a vehicle-road coordination target detection method can be implemented as a computer software program, tangibly embodied in a machine-readable medium, such as storage means 508 . In some embodiments, part or all of the computer program, via ROM 502 and/or communication means 509,
loaded and/or installed on the device 500; The computer program, when loaded into the RAM 503 and executed by the computing means 501, can perform one or more steps of the road-vehicle coordination target detection method described above. Alternatively, in other embodiments, the computing means 501 may be configured to perform the target detection method for road-vehicle coordination in any other suitable manner (eg, by firmware).

As used herein, various embodiments of the systems and techniques described above include digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs) , system-on-chip (SOC), complex programmable logic device (CPLD), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include the following. embodied in one or more computer programs, the one or more computer programs may be executed and/or interpreted in a programmable system including at least one programmable processor, which may be a dedicated or general purpose programmable a processor, receiving data and instructions from a storage system, at least one input device, and at least one output device; may be transmitted to

Preferably, in addition to including electronic devices, the roadside equipment may further include communication components and the like. Electronic devices may be integrated with communication components or configured separately. Electronic devices are capable of image video processing and data computation by acquiring data, such as pictures and videos, from sensing devices (eg, roadside cameras). Preferably, the electronic device itself may have a function of acquiring sensing data and a communication function, such as an AI camera. Electronic devices can directly perform image video processing and data calculations based on the acquired sensory data.

Preferably, the cloud control platform performs processing on the cloud side. Electronic devices included in the cloud control platform can acquire data, such as pictures and videos, from sensing devices (eg, roadside cameras) to perform image video processing and data computation. The cloud control platform may also be called a road-vehicle coordinated management platform, an edge computing platform, a cloud computing platform, a center system, a cloud-side server, and the like.

Program code for implementing the methods of the present disclosure may be compiled using any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer or other programmable data processing apparatus such that when the program code is executed by the processor or controller, it may be defined in flowchart illustrations and/or block diagrams. functions/operations can be performed. The program code can be fully machine-executed, partially machine-executed, partially machine-executed as an independent software package, and partially machine-executed and partially remote machine-executed. , or can be executed entirely on a remote machine or server.

In the context of the present disclosure, a machine-readable medium is a tangible medium that contains or stores a program with which an instruction execution system, apparatus, or device is used, or that is used in conjunction with an instruction execution system, apparatus, or device. can be A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus or devices, or any suitable combination thereof. . More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory. It may include memory (EPROM or flash memory), fiber optics, portable compact disc read only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

To provide for user interaction, the systems and techniques described above can be computer-implemented. The computer has a display device (e.g. CRT (cathode ray tube) or LCD (liquid crystal display) monitor) used to display information to the user, a keyboard and pointing device (e.g. mouse or trackball), A user can provide input information to the computer through the keyboard and the pointing device. Other types of devices can also be used to provide interaction with the user. For example, the feedback provided to the user may be any form of sensory feedback (eg, visual, auditory, or tactile feedback). It may also receive input from the user in any form (including audible, audio, and tactile input).

The systems and techniques described herein may be computing systems that include background components (e.g., as data servers), or computing systems that include intermediate components (e.g., application servers), or computing systems that include front-end components (e.g., graphical a user computer with a user interface or web browser, through which a user can interact with embodiments of the systems and techniques described herein, or such background It can be implemented in a computing system including any combination of components, intermediate components, or front-end components. The components of the system can be connected together via digital data communication (eg, a communication network) in any form or medium. Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

The computer system can include clients and servers. A client and server are generally remote from each other and interact through a communication network. A client and server relationship is established by executing computer programs on the corresponding computers and having a client-server relationship to each other. The server may be a cloud server, a distributed system server, or a blockchain-linked server.

Note that steps may be rearranged, added, or deleted in light of the various processes described above. For example, each step described in the present disclosure can be performed simultaneously, sequentially, or in any other order, and the desired result of the technical solution of the present disclosure can be obtained. There is no particular limitation in the present specification as long as it is a material.

The above specific embodiments are not considered to limit the protection scope of the present disclosure. As those skilled in the art will appreciate, various modifications, combinations, and substitutions can be made based on design requirements and other factors. Any modification, equivalent replacement, improvement, etc. made within the spirit of the present disclosure shall fall within the protection scope of the present disclosure.

Claims (23)

performing target detection on an image to obtain a candidate target region in the image, a confidence value for the candidate target region, and an occlusion degree for the candidate target region;
updating confidence values for the candidate target regions based on the IoU between the candidate target regions and the degree of occlusion of the candidate target regions;
detecting targets in the image from candidate target regions based on the updated confidence values;
A target detection method in road-vehicle cooperation. Updating a confidence value for a candidate target region based on the IoU between the candidate target regions and the degree of occlusion of the candidate target region comprises:
Select a first region with the highest confidence value from the region set, and determine the confidence of the other regions based on the IoU between the other regions in the region set and the first region and the degree of occlusion of the other regions repeating updating the degree value until the region set includes one region;
wherein the region set includes unselected regions in candidate target regions;
The method of claim 1. Updating confidence values of other regions based on IoU between other regions in the region set and the first region and degrees of occlusion of other regions,
calculating IoU between other regions in a region set and the first region;
determining a first reliability adjustment value based on the IoU and a preset IoU threshold;
Determining a second reliability adjustment value based on the degree of occlusion of other regions;
adjusting confidence values for other regions using the first confidence adjustment and the second confidence adjustment;
3. The method of claim 2. Determining a first reliability adjustment value based on the IoU and a preset IoU threshold includes:
determining whether the IoU is less than a preset IoU threshold;
determining a first reliability adjustment value as 1 if the IoU is less than a preset IoU threshold;
determining the first reliability adjustment value as the difference between 1 and the IoU if the IoU is greater than or equal to a preset IoU threshold;
4. The method of claim 3. Determining a second reliability adjustment value based on the degree of occlusion of the other region,
determining a second reliability adjustment value g(occ_pred) according to the formula:
where occ_pred is the degree of occlusion of other regions, α is a preset constant, and α
>1
4. The method of claim 3. Adjusting the reliability values of other regions using the first reliability adjustment value and the second reliability adjustment value includes:
including adjusting confidence values for other regions according to the formula:
S'=S*T1*T2
Here, S′ represents the adjusted confidence value of the other region, S represents the confidence value of the other region before adjustment, T1 represents the first confidence adjustment value, and T2 is representing the second confidence adjustment value;
A method according to any one of claims 3-5. Detecting targets in the image from candidate target regions based on the updated confidence values comprises:
selecting candidate target regions for which the updated confidence value is greater than a preset confidence threshold, and establishing targets in the selected candidate target regions as targets in the image;
or
selecting a preset number of candidate target regions with the highest updated confidence values, and establishing targets in the selected candidate target regions as targets in the image;
A method according to any one of claims 1-5. performing target detection on the image to obtain candidate target regions in the image, confidence values of the candidate target regions, and degrees of occlusion of the candidate target regions;
performing target detection on an image for each different target scale to obtain candidate target regions of different scales in the image, confidence values of the candidate target regions, and degrees of occlusion of the candidate target regions; include,
A method according to any one of claims 1-5. performing target detection on the image to obtain candidate target regions in the image, confidence values of the candidate target regions, and degrees of occlusion of the candidate target regions;
An image is input to a target detection model obtained by pre-training, and a candidate target region in the image output by the target detection model, a confidence value of the candidate target region, and an occlusion degree of the candidate target region are obtained. wherein the target detection model includes a target detection layer for detecting candidate target regions in an image and an occlusion prediction layer for predicting occlusion of candidate target regions. ,including,
A method according to any one of claims 1-5. an information acquisition module for performing target detection on an image to acquire candidate target regions in the image, confidence values for the candidate target regions, and occlusion degrees for the candidate target regions;
a confidence value update module for updating the confidence values of the candidate target regions based on the IoU between the candidate target regions and the degree of occlusion of the candidate target regions;
a target detection module for detecting targets in said image from candidate target regions based on updated confidence values;
A target detection device for road-vehicle cooperation. The confidence value update module selects a first region with the highest confidence value from a region set, and determines the IoU between other regions in the region set and the first region and the degree of occlusion of the other regions. and updating confidence values of other regions based on iterating until one region is included in the region set;
wherein the region set includes unselected regions in candidate target regions;
11. Apparatus according to claim 10. The confidence value update module includes:
IoU calculation means for calculating the IoU between other regions in the region set and the first region;
a first adjustment value determination means for determining a first reliability adjustment value based on the IoU and a preset IoU threshold;
a second adjustment value determining means for determining a second reliability adjustment value based on the degree of shielding of other regions;
reliability value adjustment means for adjusting reliability values of other regions using the first reliability adjustment value and the second reliability adjustment value;
12. Apparatus according to claim 11. The first adjustment value determination means determines whether or not the IoU is smaller than a preset IoU threshold, and sets the first reliability adjustment value to 1 if the IoU is smaller than the preset IoU threshold. and determining the first reliability adjustment value as the difference between 1 and the IoU if the IoU is greater than or equal to a preset IoU threshold.
13. Apparatus according to claim 12. The second adjustment value determination means is used to determine a second reliability adjustment value g(occ_pred) according to the following formula,
where occ_pred is the degree of occlusion of other regions, α is a preset constant, and α>1;
13. Apparatus according to claim 12. The reliability value adjusting means is used to adjust the reliability values of other regions according to the following formula:
S'=S*T1*T2
Here, S′ represents the adjusted confidence value of the other region, S represents the confidence value of the other region before adjustment, T1 represents the first confidence adjustment value, and T2 is representing the second confidence adjustment value;
Apparatus according to any one of claims 12-14. The target detection module includes:
selecting candidate target regions for which the updated confidence value is greater than a preset confidence threshold, and establishing targets in the selected candidate target regions as targets in the image;
or
selecting a preset number of candidate target regions with the highest updated confidence values, and establishing targets in the selected candidate target regions as targets in the image;
Apparatus according to any one of claims 10-14 The information acquisition module performs target detection on an image for each different target scale, and includes candidate target regions of different scales in the image, confidence values of the candidate target regions, and degrees of occlusion of the candidate target regions. used to get the
A device according to any one of claims 10-14. The information acquisition module inputs an image to a pre-trained target detection model, and extracts candidate target regions in the image output by the target detection model, confidence values of the candidate target regions, and candidate targets. It is used to obtain the degree of occlusion of the area,
The target detection model includes a target detection layer for detecting candidate target regions in an image and an occlusion prediction layer for predicting occlusion of candidate target regions,
A device according to any one of claims 10-14. at least one processor;
a memory in communication with the at least one processor;
The memory stores instructions to be executed by the at least one processor, and the execution of the instructions by the at least one processor causes the at least one processor to perform any one of claims 1 to 9. carrying out the method of one;
electronic device. A non-transitory computer-readable storage medium having computer instructions stored thereon, said computer instructions being used to cause said computer to perform the method of any one of claims 1-9,
A non-transitory computer-readable storage medium. comprising a computer program which, when executed by a processor, implements the method of any one of claims 1-9,
computer program product. Roadside equipment comprising the electronic device of claim 19 . A crowd control platform comprising the electronic device of claim 19 .