JP6847708B2 - Bed positioning device - Google Patents

Description

The present invention relates to a bed position specifying device that identifies a bed position in which a person is lying down from a captured image when the state of a person to be watched is monitored by an image captured by a camera.

In hospitals, elderly homes and facilities for the elderly, accidents due to falls of the elderly occur frequently. Since a fall accident in such a facility often occurs when getting up from the bed and getting out of bed, there is a system for watching over by installing a camera that captures a person to be watched over such as a patient on the bed. For example, in Patent Document 1, the entire bed was imaged from above the patient's head, and the patient's rising motion was discriminated from the image.

Japanese Patent No. 6046559

D. G. Lowe: “Distinctive Image Features from Scale-Invariant Keypoints”, Inter-national Journal of Computer Vision, Vol.60, No.2, pp.91-110, 2004. C. Harris and M. Stephens: “A combined corner and edge detector.” Alvey Vision Conference, pp. 147-152, 1988. N. Dalal, B. Triggs: “Histograms of Oriented Gradients for Human Detection” In IEEE Conference on Computer Vision and Pattern Recognition, vol.1, pp.886-893, 2005.

However, since the conventional technique for detecting the movement of a person such as getting up is to detect the movement of the person from the movement information of the entire imaging area, a person other than the person to be watched, such as the movement of a nurse or a visitor around the bed. In some cases, the movement of the subject was watched over and falsely detected as the movement of the subject.

Therefore, in view of such a problem, the present invention is characterized in providing a bed position specifying device that identifies a bed position from an image captured by a camera so as not to watch over a person around the bed and determine that the person is a target person.

In order to solve the above problem, the invention of claim 1 is a camera that images the entire room in which the bed is arranged from above, and an entire image captured by the camera, which has a specific shape and size for setting a bed area. An image division part that provides an overlapping part by a frame and divides it into a plurality of parts, a feature extraction part that extracts image features from individual small area images generated by the frame, and a learning sample divided into two classes depending on the presence or absence of a bed. A learning unit that stores the image features that have been created, a first determination unit that determines the presence or absence of a bed based on the image features extracted by the feature extraction unit and information from the learning unit, and a small determination unit that determines that there is a bed. A brightness calculation unit that calculates the average brightness of the entire area of the small area image and the average brightness of the internal area in which the constant peripheral width of the small area image is deleted from the area image, and the calculated total area and the internal area. It is provided with a second determination unit that determines the presence of a bed based on two average brightness information, and the second determination unit is characterized in that if the average brightness of the entire area is lower than the average brightness of the internal area, the presence of a bed is determined. To do.
According to this configuration, in order to specify the bed position in the captured image, the watching area can be limited to a specific area in the imaging area, the movement of a person around the bed can be eliminated, and false detection can be greatly reduced. Can be done. In addition, since the bed area is determined from both the image feature and the brightness feature, the bed position can be reliably specified, and the brightness calculation is performed only for the specific small area image, so that the amount of calculation can be reduced.

The invention of claim 2 is characterized in that, in the configuration according to claim 1, the frame for generating a small area image is about 10% larger than the size of the bed in the captured image.
According to this configuration, it is possible to include the entire bed in any of the plurality of generated small area images, and the small area image generated by the frame can be watched and used satisfactorily as the motion determination area of the target person. ..

According to the present invention, since the bed position in the captured image is specified, the watching area can be limited to the specified area, the movement of a person around the bed can be eliminated, and erroneous detection can be greatly reduced. In addition, since the bed area is determined from both the image feature and the brightness feature, the bed position can be reliably specified, and the brightness calculation is performed only for the specific small area image, so that the amount of calculation can be reduced.

It is a block diagram which shows an example of the bed position specifying apparatus which concerns on this invention. It is explanatory drawing of the image division processing, (a) is the appearance of dividing by a frame sufficiently large with respect to the area of the bed to image, (b) is the appearance of dividing by a frame slightly larger than a bed, (c) is a state of dividing by a frame slightly larger than a bed It shows how it is divided into slightly smaller frames. The brightness histogram for the entire area when the small area image includes the entire bed is shown, (a) is the small area image to be calculated, and (b) is the calculated brightness histogram. A luminance histogram for an internal region generated by deleting a constant peripheral width with respect to the small region image of FIG. 3 is shown, (a) an internal region to be calculated, and (b) is a calculated luminance histogram. The brightness histogram for the entire area when the small area image is the background area not including the bed is shown, (a) is the small area image to be calculated, and (b) is the calculated brightness histogram. A luminance histogram for an internal region generated by deleting a constant peripheral width with respect to the small region image of FIG. 5 is shown, (a) an internal region to be calculated, and (b) is a calculated luminance histogram.

Hereinafter, embodiments embodying the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing an example of a bed position specifying device according to the present invention, in which 1 is a camera provided with an image pickup element, 2 is an image holding unit for temporarily holding an image captured by the camera 1, and 3 is an captured image. Is an image dividing unit that divides the image into a plurality of small area images, 4 is a feature extraction unit that extracts image features for each small area image, 5 is a learning unit that stores learning data of image features, and 6 is extracted by a feature extraction unit 4. The first determination unit that determines the presence or absence of a bed based on the image features and the data of the learning unit 5, and 7 is the inside of the small area image that the first determination unit 6 determines that there is a bed, excluding the entire area and the surroundings. A brightness calculation unit that calculates the average brightness of the region, and 8 is a second determination unit that determines the presence or absence of a bed based on the calculation result of the brightness calculation unit 7.
The image holding unit 2, the image dividing unit 3, the feature extraction unit 4, the learning unit 5, the first determination unit 6, the brightness calculation unit 7, and the second determination unit 8 are integrated by the CPU or DSP in which the operation program is installed. It is composed.

The camera 1 is arranged in the upper part of the room where the patient is present and is installed so as to take an image of the entire room in order to take an image of a person to be watched over such as a patient (hereinafter, simply referred to as a "patient"). By arranging the camera 1 in this way, it is possible to fit the entire bed in the imaging area regardless of where the bed is moved in the room without moving the camera 1 or changing the angle of the camera 1. There is. Then, a moving image is output from the camera 1, and this moving image is composed of image frames (still images) that are continuously output at predetermined intervals.

The image holding unit 2 is composed of, for example, a RAM, and is primarily stored until the image (image frame) captured by the camera 1 is subjected to image division processing or the like in the subsequent stage.

The image dividing unit 3 divides the image of the image holding unit 2, that is, the image frame output by the camera 1 into a plurality of small area images. At this time, the frame set for generating the small area image has a preferable shape and size, and is set at the start of operation.
The shape is determined by the shape of the bed to be imaged, and since the camera 1 is basically fixed so as to be arranged vertically, a vertically long rectangle is selected as the frame to be used. Hereinafter, this shape will be described.

Regarding the size, frames having a plurality of different sizes are prepared, and FIG. 2 shows the concept of image division using frames having different sizes. FIG. 2A shows a state in which the area of the bed to be imaged is divided into a sufficiently large frame, FIG. 2B shows a state in which the area is divided into a frame slightly larger than the bed, and FIG. 2C shows a state in which the area is slightly smaller than the bed. The state of dividing by a frame is shown, and the state of dividing into small areas of three different sizes is shown.

In the generation of the small area image, the small area image is sequentially generated and generated by moving the frame in parallel so as to have an overlapping portion and divide the entire image with respect to the image frame output by the camera 1. The small area image data is output in the subsequent stage, and image processing is sequentially performed on the output images.

The bed size in the image frame is determined to be almost constant with respect to the image size, but since the bed size differs depending on the distance between the bed and the camera, a plurality of types of sizes are prepared as shown in FIG. , It is possible to select the optimum frame size with a simple operation by the selection operation without making a new setting.

At that time, the frame size is selected based on the captured image when the position of the bed is determined, and the size is equal to or close to the size of the bed to be imaged with a margin of about 10% in width and height. The frame is selected. By providing a margin of about 10%, any small area image becomes an image including the entire bed or almost the entire bed, so that the bed can be easily captured from the image. Further, by setting the small area itself as the bed area and excluding the surrounding area from the judgment target, it is easy to exclude the surrounding persons other than the patient to be watched when recognizing the movement of the patient.

The feature extraction unit 4 extracts image features from the small area image. Image features are disclosed in, for example, SIFT disclosed in Non-Patent Document 1, Harris corners representing corners, for example, as disclosed in Non-Patent Document 2, or, for example, Non-Patent Document 3 representing the gradient of an image. An image feature expressing the shape of an object is calculated (extracted) by applying a HOG or the like as described above.
This extraction is sequentially performed on each divided subregion image.

The learning unit 5 arranges beds in various places in the room and classifies the data into a plurality of captured images obtained by capturing the entire room from above by the camera 1 in the above-arranged state and a captured image without a bed. The feature amount information calculated based on the basis is stored.

The first determination unit 6 is composed of a learning sample stored by the learning unit 5 divided into two classes, a bed and a non-bed, and a strong classifier learned using the feature amount calculated by the feature extraction unit 4. Using the well-known Adaboost classifier, it is determined whether or not the small area image divided by the image dividing unit 3 is a bed (whether or not there is a bed).
Various methods such as SVM, subspace method, and neural network can be applied to the classifier.

The brightness calculation unit 7 calculates the average brightness for the entire area and the average brightness for the internal area with respect to the small area image determined by the first determination unit 6 to have a bed. As described above, the small area image divided by the image dividing unit 3 is generated by using a frame having a size obtained by adding a margin of about 10% to the size of the bed. Therefore, the average brightness is calculated with respect to the internal region, with the region in which 10% of the periphery of the small region image is deleted as the internal region.

3 and 4 show a luminance histogram when the small area image includes the entire bed, FIG. 3 shows the entire area, and FIG. 4 shows the internal area. In each case, (a) shows a small area image to be calculated, and (b) is a luminance histogram calculated. It can be seen that the inside of the frame shown in FIG. 4A is the internal region, and the histogram for the internal region has a smaller dark region than the histogram for the entire region (the score in the low-luminance region is low).

Here, the luminance histograms calculated for the small area image not including the bed as a comparison target are shown in FIGS. 5 and 6. FIG. 5 shows a brightness histogram for the entire area when the small area image is a background area that does not include a bed, (a) is a small area image to be calculated, and (b) is a calculated brightness histogram. FIG. 6 shows a luminance histogram for an internal region generated by deleting a constant peripheral width with respect to the small region image of FIG. 5, (a) an internal region to be calculated, and (b) is a calculated luminance histogram.
As shown in FIGS. 5 and 6, in the case of the small area image not including the bed, the brightness histogram obtained from the internal area is higher brightness information around the outer circumference than the brightness histogram obtained from the entire area of the small area image. It can be seen that does not include (the score in the high brightness region is low).

The second determination unit 8 determines the bed from the two average brightnesses obtained from the brightness calculation unit 7. The average brightness of the entire region and the internal region of the small region image is compared, and if the average brightness of the entire region is lower than the average brightness of the internal region, it is determined that the divided region is a bed.
As described above, when the small area image to be processed corresponds to the bed area, the luminance histogram obtained from the internal region of the small area image has a lower score in the low brightness region than the luminance histogram obtained from all the regions. On the contrary, in the case of the small area image not including the bed, the luminance histogram obtained from the internal region of the small area image has a lower score in the high luminance region than the luminance histogram obtained from all the regions. From this, the average brightness of the entire region and the internal region of the small region image is compared, and if the average brightness of the entire region is lower than the average brightness of the internal region, it is determined that the region is a bed.
Then, the motion determination of the person is subsequently made only for the region of the determined small area image (not described in detail).

In this way, in order to specify the bed position in the captured image, the watching area can be limited to a specific area in the imaging area, the movement of a person around the bed can be eliminated, and false detection can be greatly reduced. it can. In addition, since the bed area is determined from both the image feature and the brightness feature, the bed position can be reliably specified, and the brightness calculation is performed only for the specific small area image, so that the amount of calculation can be reduced.
Furthermore, by generating a small area image with a frame that is about 10% larger than the bed, it is possible to include the entire bed in any of the plurality of generated small area images, and the small area image generated by the frame. It can be satisfactorily used as an operation judgment area for the target person while watching over.

In the above embodiment, the margin with respect to the bed size of the frame for generating the small area image is set to about 10%, but the movement of the patient can be determined even if the margin is set to about the same as the size of the bed in the image frame. However, a margin of about 20% from the bed size may be provided. If the margin is large, the probability of determining the person to be watched, including the people around the patient other than the patient on the bed, increases, but if it is about 20%, the probability can be suppressed to a low level.
However, in that case, it is desirable to increase the amount of deletion of the surroundings to about 20% when calculating the average brightness of the internal region.
Further, the frame for generating the small area image may be square or horizontally long, and when the bed is arranged horizontally in the image captured by the camera 1, it is preferable to prepare a horizontally long frame.

1 ... camera, 2 ... image holding unit, 3 ... image dividing unit, 4 ... feature extraction unit, 5 ... learning unit, 6 ... 1st judgment unit, 7 ... brightness calculation unit, 8 ... Second judgment unit.

Claims (2)

A camera that captures the entire room where the bed is placed from above,
An image dividing portion that divides the entire image captured by the camera into a plurality of images by providing overlapping portions according to a frame having a specific shape and size for setting a bed area.
A feature extraction unit that extracts image features from individual small area images generated by the frame, and
A learning unit that stores the image features extracted from the learning samples divided into two classes according to the presence or absence of a bed, and
A first determination unit that determines the presence or absence of the bed based on the image features extracted by the feature extraction unit and the information of the learning unit.
For the small area image determined by the first determination unit to have a bed, the average brightness of the entire area of the small area image and the average brightness of the internal area in which the constant peripheral width of the small area image is deleted are calculated. Brightness calculation unit and
It is provided with a second determination unit for determining the presence or absence of a bed based on the calculated two average brightness information of the entire area and the internal area.
The second determination unit is a bed position specifying device, characterized in that if the average brightness of the entire area is lower than the average brightness of the internal area, it is determined that there is a bed. The bed position specifying device according to claim 1, wherein the frame for generating the small area image is about 10% larger than the size of the bed in the captured image.

Images