JP7706975B2 - DETECTION DEVICE, DETECTION METHOD, AND DETECTION PROGRAM - Google Patents

Description

The present invention relates to a detection device, a detection method, and a detection program.

Conventionally, there are known techniques for detecting objects in an image. For example, there is known a technique for extracting an area in an image in which a human body appears (see, for example, Patent Document 1).

JP 2008-15641 A

However, conventional technology has the problem that it may not be possible to accurately detect a human face from an image.

For example, the technology described in Patent Document 1 detects a human body based on the detection results of a human face. Here, in detecting an object, the detection accuracy may decrease as the size of the object becomes smaller. Therefore, the technology described in Patent Document 1 may not be able to detect a human face, and as a result, may not be able to detect a human body.

The present invention has been made in consideration of the above, and aims to accurately detect human faces from images.

In order to solve the above-mentioned problems and achieve the object, the detection device according to the present invention has a detection unit and a determination unit. The detection unit detects an area including a human face and an area including a human body from an image, together with the degree of certainty for each. The determination unit determines whether face detection has been successful or not, based on the degree of certainty corresponding to the area including the face and the degree of certainty corresponding to the area including the body.

The present invention makes it possible to accurately detect human faces from images.

FIG. 1 is a diagram illustrating an example of the configuration of a detection device according to an embodiment. FIG. 2 is a diagram showing an example of an image and a detection result. FIG. 3 is a diagram illustrating an example of the detection rule. FIG. 4 is a diagram showing an example of an image and a detection result. FIG. 5 is a flowchart showing a process flow of the detection device.

Below, the embodiments of the detection device, detection method, and detection program disclosed in the present application will be described in detail with reference to the attached drawings. Note that the present invention is not limited to the embodiments described below.

First, a detection device according to an embodiment will be described with reference to FIG. 1. FIG. 1 is a diagram showing an example of the configuration of a detection device according to an embodiment.

Here, in order to protect personal information, certain objects in an image may be concealed. For example, concealment is performed by filling in or mosaic processing the area in the image in which the target object appears.

For example, people's faces and license plates of other vehicles may be concealed in images captured by a drive recorder installed in a vehicle.

The detection device 10 detects a human face from the input image. For example, the detection result by the detection device 10 may be used to conceal a human face in the image.

The detection device 10 receives input of images captured by an in-vehicle device such as a drive recorder mounted on a vehicle. The images captured by the in-vehicle device may include the face of a pedestrian.

The detection device 10 may be realized as part of a function of an in-vehicle device. The detection device 10 may also be an information processing device such as a personal computer or a server.

The detection device 10 can detect an area in which an object is captured along with the degree of likelihood, for example, by using a detection method that uses deep learning. For example, the detection device 10 detects a rectangular area called a bounding box.

Furthermore, the detection device 10 determines whether the detection result of the object satisfies a condition. In this embodiment, if the detection device 10 determines that the detection result satisfies a condition, it is deemed that the object corresponding to the detection result has been detected.

The degree of certainty may be the probability that an object exists in the area, or a value that increases as the probability increases. In this embodiment, the degree of certainty detected by the detection device 10 is called the confidence level.

As shown in FIG. 1, the detection device 10 has an interface unit 11, a memory unit 12, and a control unit 13.

The interface unit 11 is an interface for communicating data with other devices. The detection device 10 accepts image input via the interface unit 11. The detection device 10 also outputs detection results via the interface unit 11.

The control unit 13 and memory unit 12 of the detection device 10 are realized by a computer having, for example, a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), flash memory, input/output ports, etc., and various other circuits.

The storage unit 12 corresponds to a RAM or a flash memory. The RAM or the flash memory stores the model information 121.

Model information 121 is information for constructing a model to detect objects from an image.

Here, the model constructed by the model information 121 is, for example, a trained neural network. In this case, the model information 121 is parameters such as weights and biases for constructing the neural network.

The computer's CPU functions as the detection unit 131 and determination unit 132 of the control unit 13, for example, by reading and executing a program stored in the ROM.

The detection device 10 may also acquire the above-mentioned programs and various information via other computers or portable recording media connected via a wired or wireless network.

The detection unit 131 detects areas including a human face and areas including a human body from an image, along with their respective confidence levels (degrees of certainty).

The detection unit 131 detects an object by a predetermined method using a model such as a neural network constructed from the model information 121. For example, the detection device 10 may detect an object by a known method using deep learning such as YOLO.

In this embodiment, the body refers to the entire body, including the face.

The determination unit 132 determines whether or not a human face has been detected based on the confidence level corresponding to the area including the face and the confidence level corresponding to the area including the body.

For example, if the confidence level corresponding to an area detected to contain a human face is equal to or greater than a threshold, the determination unit 132 determines that a human face has been detected in that area.

Here, when determining whether a human face has been detected in an image, the determination unit 132 refers to the detection results in the area including the body as well as the detection results in the area including the face.

This is because the size of a human face is small compared to the entire body, which can reduce detection accuracy. For example, detection misses are more likely to occur when the confidence level is close to the threshold.

In addition to the small size of human faces, a decrease in detection accuracy can also be caused by blurring of images due to vehicle speed (motion blur, etc.), blurring of objects due to rain or fog, and blurring due to camera lens distortion.

The following describes an example of the detection device 10.

Example 1
The determination unit 132 can determine that a human face has been detected when the certainty factor corresponding to the region including a face is less than a threshold value and the certainty factor corresponding to the region including a body is equal to or greater than the threshold value.

This allows the body detection results to be used to detect a human face even when the face cannot be detected with sufficient confidence.

Figure 2 shows an example of an image and a detection result. Region 201 and region 202 are regions detected by the detection unit 131. Region 201 is a region that includes a face. Region 202 is a region that includes a body.

Here, the confidence threshold is set to 0.3. The confidence of the face in region 201 is set to 0.1. The confidence of the body in region 202 is set to 0.6.

In this case, the certainty of the face in region 201 is less than the threshold. On the other hand, the certainty of the body in region 202 is equal to or greater than the threshold. As a result, the determination unit 132 determines that a face has been detected in region 201 for the image in FIG. 2.

Example 2
The determination unit 132 can make a determination based on a plurality of rules set in advance. Fig. 3 is a diagram showing an example of the detection rules. The rules shown in Fig. 3 may be stored in the storage unit 12.

"Face detection" in Figure 3 is the detection result of an area including a face by the detection unit 131.

The "Face Detection" column indicates whether the confidence level of the area containing a face detected by the detection unit 131 is equal to or greater than a threshold (detected) or is less than the threshold (not detected).

Furthermore, the "Face Detection" column indicates that the confidence level is 0. A confidence level of 0 means that the detection unit 131 did not detect an area containing a face in the first place. Note that the threshold value is greater than 0.

"Human detection" in Figure 3 is the detection result of an area including a human body by the detection unit 131.

The "Human Detection" column indicates whether the confidence level of the area containing the body detected by the detection unit 131 is equal to or greater than a threshold (detected) or is less than the threshold (not detected).

Furthermore, the "Human Detection" column indicates that the confidence level is 0. A confidence level of 0 means that the detection unit 131 did not detect an area that includes a body in the first place.

The "face + person determination" in FIG. 3 is the determination result by the determination unit 132. The determination unit 132 compares the detection result by the detection unit 131 with the "face detection" and "person detection" and outputs the content of the corresponding "face + person determination" (detected or not detected) as the determination result.

Note that "face only judgement" is the judgement result when the detection result is judged by referring only to "face detection".

First, if the confidence level of "face detection" is equal to or greater than a threshold, the determination unit 132 determines that a face has been detected (No. 1, 4, and 5 in Figure 3).

In addition, if the confidence level of "face detection" is 0, the determination unit 132 determines that a face has not been detected (undetected) (No. 3, 7, 8 in Figure 3).

Here, if the confidence level of "face detection" is less than the threshold and not 0, and the confidence level of "person detection" is equal to or greater than the threshold, the determination unit 132 determines that a face has been detected (No. 2 and 6 in Figure 3).

As shown in Figure 3, for No. 2 and 6, the results of "face + person determination" are different from the results of "face only determination."

Note that No. 3 corresponds to a situation where, for example, the body below the neck is visible in the image, but the face is not visible due to posture or an obstruction.

And No. 4 corresponds to a situation where, for example, the face is in the image, but other body parts are hidden.

In addition, No. 5 corresponds to the opposite of No. 3, where, for example, the face is in the image but the person's head and below are not visible.

Note that No. 8 corresponds to a situation where the face is not visible in the image due to posture or an obstruction, for example, and part of the body below the neck is also hidden.

Example 3
When the detection unit 131 detects multiple areas including a human face and one area including a human body, the judgment unit 132 can judge whether or not the detection of a human face has been successful for the area among the multiple areas including the face that has the largest overlapping area with the area including the body.

This makes it possible to match the face and body of the same person in an image, even when people are crowded together.

Figure 4 is a diagram showing an example of an image and a detection result. Regions 211, 212, and 213 are regions detected by the detection unit 131. Regions 211 and 212 are regions that include a face. Region 213 is a region that includes a body.

Before making a detection determination, the determination unit 132 determines whether the face area corresponding to the body detected in area 213 is area 211 or area 212.

At this time, since the area of the overlapping area between areas 212 and 213 is larger than the area of the overlapping area between areas 211 and 213, the determination unit 132 determines that the face area corresponding to the body detected in area 213 is area 212.

In the example of FIG. 4, the entire area of region 212 overlaps with region 213, whereas a portion of region 211 overlaps with region 213. In this way, the determination unit 132 may identify a region that entirely overlaps with a body region as the corresponding face region.

Example 4
The determination unit 132 can determine that a human face has been detected if the confidence level corresponding to the area including the face is less than a first threshold and the confidence level corresponding to the area including the body is equal to or greater than a second threshold that is greater than the first threshold.

This improves detection accuracy while reducing the occurrence of false positives caused by lowering the threshold.

The determination unit 132 changes the threshold value depending on the object (face or body) to be detected. In FIG. 2, the certainty factor of the face in region 201 is assumed to be 0.1. The certainty factor of the body in region 202 is assumed to be 0.6.

At this time, the determination unit 132 sets the first threshold to 0.1. The determination unit 132 also sets the second threshold to 0.3.

In this case, the certainty of the body in region 202 is equal to or greater than the second threshold. Furthermore, the certainty of the face in region 201 is equal to or greater than the first threshold. Furthermore, as a result, the determination unit 132 determines that a face has been detected in region 201 for the image in FIG. 2.

Note that the determination unit 132 sets the first threshold to a value smaller than the reference value of 0.3, but conversely, the second threshold may be set to a value larger than the reference value of 0.3.

Figure 5 is a flowchart showing the flow of processing by the detection device. As shown in Figure 5, the detection device 10 detects an area including a human face from an image (step S101). The detection device 10 also detects an area including a human body from an image (step S102).

Here, the detection device 10 determines whether the certainty of each detection result satisfies the condition (step S103).

The conditions are as described in each of the above embodiments. For example, the detection device 10 determines that the conditions are met if the body certainty in the detection result is equal to or greater than a threshold, and the face certainty is less than the threshold and not 0.

If the confidence level for each detection result satisfies the condition (step S103, Yes), the detection device 10 determines that a face has been detected (step S104).

On the other hand, if the confidence level for each detection result does not satisfy the condition (step S103, No), the detection device 10 determines that the face has not been detected (step S105).

As described above, the detection device 10 according to the embodiment has a detection unit 131 and a determination unit 132. The detection unit 131 detects an area including a human face and an area including a human body from an image, together with a certainty factor (degree of certainty) for each. The determination unit 132 determines whether face detection has been successful, based on the certainty factor corresponding to the area including the face and the certainty factor corresponding to the area including the body.

In this way, the detection device 10 refers to not only the detection results related to a person's face, but also the detection results related to the body to determine whether or not a face has been detected. Therefore, according to this embodiment, it is possible to accurately detect a person's face from an image.

Further advantages and modifications may readily occur to those skilled in the art. Therefore, the invention in its broader aspects is not limited to the specific details and representative embodiments shown and described above. Accordingly, various modifications may be made without departing from the spirit or scope of the general inventive concept as defined by the appended claims and equivalents thereof.

Objects that are detected and concealed to protect personal information are not limited to human faces. For example, vehicle license plates may appear in images captured by a drive recorder. Here, the information written on a vehicle license plate can also be considered personal information and may be subject to concealment.

The embodiment can be applied not only to human faces but also to license plates. That is, the detection device 10 detects an area including a license plate and an area including a vehicle from an image, together with a certainty factor for each. Then, the detection device 10 determines whether or not the license plate has been detected based on the certainty factor corresponding to the area including the license plate and the certainty factor corresponding to the area including the vehicle.

REFERENCE SIGNS LIST 10 Detection device 11 Interface unit 12 Storage unit 13 Control unit 121 Model information 131 Detection unit 132 Determination unit 201, 202, 211, 212, 213 Area

Claims (4)

a detection unit that detects an area including a human face and an area including a human body from an image together with a degree of likelihood of each area;
a determination unit that determines that a human face has been detected when a degree of certainty corresponding to the region including the face is less than a first threshold and a degree of certainty corresponding to the region including the body is equal to or greater than a second threshold that is greater than the first threshold ;
A detection device comprising: The detection device according to claim 1, characterized in that when the detection unit detects multiple areas including a human face and one area including a human body, the determination unit determines whether or not a human face has been detected for the area among the multiple areas including the face that has the largest overlapping area with the area including the body. A detection method performed by a detection device, comprising:
a detection step of detecting an area including a human face and an area including a human body from the image together with a degree of likelihood of each area;
a determining step of determining that a human face has been detected when a degree of certainty corresponding to the region including the face is less than a first threshold and a degree of certainty corresponding to the region including the body is equal to or greater than a second threshold that is greater than the first threshold ;
A detection method comprising: On the computer,
a detection step of detecting an area including a human face and an area including a human body from an image, each of the areas being detected together with a degree of likelihood;
a determination step of determining that a human face has been detected if a degree of certainty corresponding to the region including the face is less than a first threshold and a degree of certainty corresponding to the region including the body is equal to or greater than a second threshold that is greater than the first threshold ;
A detection program characterized by executing the above.

Images