Fall detection optimization method and system based on infrared image features

Abstract

This invention discloses an optimized method and system for fall detection based on infrared image features, belonging to the field of computer vision and intelligent monitoring technology. It addresses the problem of poor accuracy and robustness in fall detection under low light conditions. The method includes: acquiring infrared fall images from multiple scenes, constructing a dataset after bilateral filtering for denoising and histogram equalization enhancement; improving the YOLOv11n network by introducing the MPDIoU loss function to optimize bounding box regression accuracy, and replacing the Conv module with a lightweight adaptive extraction (LAE) module to obtain a target detection model; inputting real-time infrared images into the target detection model and outputting fall detection results. This invention, through the collaborative design of infrared image preprocessing, the MPDIoU loss function, and the lightweight LAE module, significantly improves the accuracy and robustness of fall detection in low light conditions while maintaining model real-time performance, providing reliable technical support for intelligent monitoring of the elderly under complex conditions such as nighttime and shadows.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and intelligent monitoring technology, and in particular relates to an optimized method and system for fall detection based on infrared image features. Background Technology

[0002] Traditional fall detection methods are mainly divided into three categories: wearable sensor-based methods, environmental perception-based methods, and computer vision-based methods. Wearable devices (such as accelerometers and gyroscopes) need to be worn all over the body, which has problems such as poor comfort, frequent charging, and easy forgetting. Environmental perception methods (such as infrared beam sensors and pressure sensors) have high deployment costs and limited coverage. Computer vision-based methods, with their advantages of being non-contact, providing rich information, and being real-time, are gradually becoming the mainstream research direction in the field of fall detection.

[0003] However, fall detection faces severe challenges in real-world applications: First, indoor and outdoor lighting conditions are complex and variable. In low-light environments such as nighttime, cloudy days, and shadows, the quality of visible light images drops sharply, blurring human outlines and significantly reducing detection accuracy. Second, fall postures are diverse, including forward leaning, side falls, backward leaning, and slow slips, which are visually highly similar to everyday behaviors such as bending over, squatting, and lying down, easily leading to false detections. Third, when the monitoring angle is far away, the human body occupies a small portion of the image, making it easy to lose feature information. In addition, edge devices have strict limitations on the computational efficiency and parameter count of models, making it difficult to deploy complex, large models.

[0004] Infrared thermal imaging technology offers a novel approach to fall detection in low-light environments due to its advantages such as being unaffected by illumination, penetrating smoke, and clearly revealing the body's temperature distribution. However, infrared images suffer from limited detail, low contrast, and high noise, making direct application to conventional target detection models ineffective. YOLOv11n, the latest lightweight model in the YOLO series, boasts advantages such as high detection speed and a small parameter set, but it still faces challenges in infrared image processing, including insufficient feature extraction capabilities and limited bounding box regression accuracy.

[0005] Therefore, there is an urgent need for a dedicated method that can deeply integrate infrared image features, specifically improve the YOLOv11n network structure, and significantly enhance the accuracy and robustness of fall detection in low-light environments while maintaining the advantage of lightweight design. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide an optimized method and system for fall detection based on infrared image features. By preprocessing infrared images, introducing the MPDIoU loss function, and designing a lightweight adaptive extraction LAE module, high-precision, real-time detection of fall behavior in the elderly under low light conditions can be achieved.

[0007] In a first aspect, the present invention provides an optimized method for fall detection based on infrared image features, comprising: Infrared fall images in multiple scenarios are acquired, and each infrared fall image is standardized, labeled, and augmented to obtain a target infrared fall image dataset. An improved YOLOv11n network model is constructed, and the improved YOLOv11n network model is iteratively trained and validated based on the target infrared fall image dataset to obtain a target detection model. The improved YOLOv11n network model includes the introduction of the MPDIoU loss function and the replacement of the Conv module in the improved YOLOv11n network model with a lightweight adaptive extraction LAE module. The real-time infrared fall image to be detected is input into the target detection model, and the target detection model outputs the fall detection result corresponding to the real-time infrared fall image.

[0008] Secondly, the present invention provides a fall detection optimization system based on infrared image features, comprising: The acquisition module is configured to acquire infrared fall images in multiple scenarios, and to perform standardized annotation and data augmentation on each infrared fall image to obtain the target infrared fall image dataset; The module is configured to build an improved YOLOv11n network model and iteratively train and validate the improved YOLOv11n network model based on the target infrared fall image dataset to obtain a target detection model. The improved YOLOv11n network model includes the introduction of the MPDIoU loss function and the replacement of the Conv module in the improved YOLOv11n network model with a lightweight adaptive extraction LAE module. The output module is configured to input the real-time infrared fall image to be detected into the target detection model, and the target detection model outputs the fall detection result corresponding to the real-time infrared fall image.

[0009] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the fall detection optimization method based on infrared image features according to any embodiment of the present invention.

[0010] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the fall detection optimization method based on infrared image features according to any embodiment of the present invention.

[0011] This application presents an optimized method and system for fall detection based on infrared image features. The method acquires infrared fall images from multiple scenes, and constructs a dataset after bilateral filtering for denoising and histogram equalization enhancement. It improves the YOLOv11n network by introducing the MPDIoU loss function to optimize bounding box regression accuracy and replaces the Conv module with a lightweight adaptive extraction LAE module to obtain the target detection model. Real-time infrared images are input into the target detection model, and fall detection results are output. Through the collaborative design of infrared image preprocessing, the MPDIoU loss function, and the lightweight LAE module, the accuracy and robustness of fall detection in low-light environments are significantly improved while maintaining the model's real-time performance. This provides reliable technical support for intelligent monitoring of the elderly under complex conditions such as nighttime and shadows. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating an optimized fall detection method based on infrared image features, provided in an embodiment of the present invention; Figure 2 This is a structural block diagram of a fall detection optimization system based on infrared image features, provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Please see Figure 1 The diagram shows a flowchart of an optimized fall detection method based on infrared image features according to this application.

[0016] like Figure 1 As shown, the fall detection optimization method based on infrared image features specifically includes the following steps: Step S101: Acquire infrared fall images in multiple scenarios, and perform standardized annotation and data augmentation on each infrared fall image to obtain the target infrared fall image dataset.

[0017] In this step, each infrared fall image is preprocessed for the first time according to the preset bilateral filtering algorithm, and then preprocessed for the first infrared fall image according to the histogram equalization strategy to obtain each initial infrared fall image. The initial infrared fall images are labeled using an image annotation tool to obtain target infrared fall images containing the bounding rectangle of the human body. The target infrared fall images are then divided into the same dataset to obtain the target infrared fall image dataset.

[0018] Specifically, complex scenes such as indoor corridors, park squares, and outdoor lawns were selected, and infrared thermal imaging cameras were used to collect infrared images of human falling postures. The image resolution was uniformly adjusted to 640×640 pixels. The collection environment covered complex conditions such as low light at dusk, no light at night, and weak light on cloudy days. A total of no less than 2,000 falling posture samples were collected.

[0019] To effectively improve the image feature capture capability and detection robustness of subsequent object detection models, image data preprocessing is performed. First, a bilateral filtering algorithm is used to eliminate image noise, its expression being: , , , , In the formula, Let x be the pixel value of the filtered image at position x. Let be the pixel value of the original image at position y. This represents the pixel value of the original image at position x. This is the support region of the filter. Let x be the Euclidean distance between pixels x and y. For pixels With pixel value The absolute value of the difference between the pixel values, As a normalization factor, ensure that the sum of the weights is 1. This is a spatial weighting function that represents the influence of spatial distance between pixels. This is the intensity weighting function, representing the impact of differences between pixel values. For spatial standard deviation, denoted as the standard deviation of the strength.

[0020] Subsequently, histogram equalization is used to enhance the grayscale contrast of the infrared image, highlighting the human body contours and posture features. The expression for this is: , , , In the formula, For the first part of the original image The probability density of grayscale values. The grayscale value in the original image The number of samples, The total number of pixels in the image. The total number of gray levels. For the first part of the original image The cumulative distribution function of grayscale values, with values ​​ranging from [0,1]. Let i be the probability density of the i-th gray level value in the original image. The gray levels mapped after equalization. To round to the nearest integer, For the first part of the original image grayscale value, The i-th gray level value of the original image. The grayscale value in the original image The number of samples.

[0021] After preprocessing, the LabelImg image annotation tool was used to annotate the images. Bounding rectangles were used to annotate the images of the fallen postures. The annotation format was a YOLO-compatible txt format, including the target category, the center coordinates of the bounding box (x, y), and the width and height of the bounding box (w, h). All annotation data underwent manual review and correction to ensure an accuracy rate of ≥99%.

[0022] Finally, the labeled dataset was divided into training, validation and test sets in a ratio of 7:2:1. The division process adopted stratified sampling to ensure the consistency of scene and pose distribution within each dataset.

[0023] Step S102: Construct an improved YOLOv11n network model, and iteratively train and validate the improved YOLOv11n network model based on the target infrared fall image dataset to obtain a target detection model. The improved YOLOv11n network model includes the introduction of the MPDIoU loss function, and the replacement of the Conv module in the improved YOLOv11n network model with a lightweight adaptive extraction LAE module.

[0024] In this step, the MPDIoU loss function is introduced to optimize the bounding box regression accuracy. The IoU loss in the original loss function is replaced with the MPDIoU loss function, which simultaneously considers the overlap of the bounding boxes, the distance between center points, the aspect ratio, and the perimeter constraint. Its calculation formula is as follows: , , , , , , In the formula, Let MPDIoU be the loss function. The intersection-union ratio (IU) of the predicted bounding box and the ground truth bounding box is an index that combines position and shape constraints. To predict the overlap index between the bounding box and the ground truth bounding box, Let be the intersection area of ​​the two bounding boxes. Let the area be the union of the two bounding boxes. To predict the area of ​​the frame, This represents the actual area of ​​the frame. It is the square of the Euclidean distance between a diagonal vertex of the predicted bounding box and the corresponding vertex of the ground truth bounding box. It is the square of the Euclidean distance between the other diagonal vertex of the predicted bounding box and the corresponding vertex of the ground truth bounding box. , These are the coordinates of the two diagonal vertices of the prediction bounding box. , These are the coordinates of the two diagonal vertices of the real bounding box. , These are the width and height of the input image, respectively.

[0025] The traditional Conv module in the network head section is replaced with a lightweight adaptive extraction LAE module. The processing flow includes: Lightweight feature extraction is performed through depthwise separable convolution to reduce computational complexity. Feature distribution is optimized and nonlinear expressive power is enhanced by batch normalization (BN) and Hardswish activation function. By compressing the feature dimension through global average pooling (GAP) and combining fully connected layers with Sigmoid activation to generate channel attention weights, the optimized feature map is adaptively weighted to enhance the response of key features and suppress redundant information.

[0026] The expression for the LAE module is: , , , , In the formula, This is the final output feature map of the LAE module. The output feature map of depthwise separable convolution. For channel attention weights, For the input feature map, This is a global average pooling operation. This is the weight matrix of the first fully connected layer. This is the weight matrix of the second fully connected layer. for Activation function for Activation function For depthwise separable convolution operations, For batch normalization operations, For activation function, This is the feature map after channel attention weighting.

[0027] Next, the improved model is trained and validated based on the dataset constructed in step S101. Specifically, a hardware and software training environment based on the PyTorch framework is set up, and key hyperparameters such as an initial learning rate of 0.01, a batch size of 16, and 300 iterations are set. The training set is input into the improved YOLOv11n model, and data augmentation strategies such as random horizontal flipping, random cropping, and color dithering are used to improve the model's generalization ability. Iterative training is performed using the SGD optimizer, and the model performance is monitored in real time using the validation set during training to adjust the training parameters to avoid overfitting.

[0028] Step S103: Input the real-time infrared fall image to be detected into the target detection model, and the target detection model outputs the fall detection result corresponding to the real-time infrared fall image.

[0029] In summary, the method of this application firstly addresses the problems of high noise and low contrast in infrared images by employing a dual preprocessing strategy combining bilateral filtering and histogram equalization. This effectively removes image noise while enhancing human contour and pose features, providing high-quality input for subsequent detection. Secondly, it introduces the MPDIoU loss function to replace the traditional IoU loss. By comprehensively considering multi-dimensional geometric constraints such as intersection-union ratio, center point distance, aspect ratio, and perimeter constraints, it significantly improves the localization accuracy of bounding boxes for falling targets with variable poses. Thirdly, it designs a lightweight adaptive extraction LAE module to replace the traditional Conv module in the YOLOv11n network head. It utilizes depthwise separable convolution to reduce computational complexity and combines channel attention mechanism to adaptively enhance key feature responses and suppress redundant information, significantly improving feature representation capabilities while maintaining the model's lightweight advantage.

[0030] Please see Figure 2 The diagram shows a structural block diagram of a fall detection optimization system based on infrared image features according to this application.

[0031] like Figure 2 As shown, the fall detection optimization system 200 includes an acquisition module 210, a construction module 220, and an output module 230.

[0032] The acquisition module 210 is configured to acquire infrared fall images in multiple scenarios, and perform standardized annotation and data augmentation on each infrared fall image to obtain a target infrared fall image dataset; the construction module 220 is configured to construct an improved YOLOv11n network model, and perform iterative training and validation on the improved YOLOv11n network model based on the target infrared fall image dataset to obtain a target detection model, wherein the improved YOLOv11n network model includes the introduction of the MPDIoU loss function, and the Conv module in the improved YOLOv11n network model is replaced with a lightweight adaptive extraction LAE module; the output module 230 is configured to input the real-time infrared fall image to be detected into the target detection model, and the target detection model outputs the fall detection result corresponding to the real-time infrared fall image.

[0033] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0034] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the fall detection optimization method based on infrared image features in any of the above method embodiments. In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows: Infrared fall images in multiple scenarios are acquired, and each infrared fall image is standardized, labeled, and augmented to obtain a target infrared fall image dataset. An improved YOLOv11n network model is constructed, and the improved YOLOv11n network model is iteratively trained and validated based on the target infrared fall image dataset to obtain a target detection model. The improved YOLOv11n network model includes the introduction of the MPDIoU loss function and the replacement of the Conv module in the improved YOLOv11n network model with a lightweight adaptive extraction LAE module. The real-time infrared fall image to be detected is input into the target detection model, and the target detection model outputs the fall detection result corresponding to the real-time infrared fall image.

[0035] Computer-readable storage media may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application program required for at least one function; the data storage area may store data created based on the use of the infrared image feature-based fall detection optimization system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely configured relative to a processor, which can be connected to the infrared image feature-based fall detection optimization system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0036] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the fall detection optimization method based on infrared image features described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the fall detection optimization system based on infrared image features. The output device 340 may include a display screen or other display device.

[0037] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0038] In one implementation, the above-described electronic device is applied to a fall detection optimization system based on infrared image features, for a client, and includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: Infrared fall images in multiple scenarios are acquired, and each infrared fall image is standardized, labeled, and augmented to obtain a target infrared fall image dataset. An improved YOLOv11n network model is constructed, and the improved YOLOv11n network model is iteratively trained and validated based on the target infrared fall image dataset to obtain a target detection model. The improved YOLOv11n network model includes the introduction of the MPDIoU loss function and the replacement of the Conv module in the improved YOLOv11n network model with a lightweight adaptive extraction LAE module. The real-time infrared fall image to be detected is input into the target detection model, and the target detection model outputs the fall detection result corresponding to the real-time infrared fall image.

[0039] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fall detection optimization method based on infrared image features, characterized in that, include: Infrared fall images in multiple scenarios are acquired, and each infrared fall image is standardized, labeled, and augmented to obtain a target infrared fall image dataset. An improved YOLOv11n network model is constructed, and the improved YOLOv11n network model is iteratively trained and validated based on the target infrared fall image dataset to obtain a target detection model. The improved YOLOv11n network model includes the introduction of the MPDIoU loss function and the replacement of the Conv module in the improved YOLOv11n network model with a lightweight adaptive extraction LAE module. The real-time infrared fall image to be detected is input into the target detection model, and the target detection model outputs the fall detection result corresponding to the real-time infrared fall image.

2. The method of claim 1, wherein, The standardization and data augmentation of each infrared fall image to obtain the target infrared fall image dataset includes: The infrared fall images are preprocessed for the first time according to the preset bilateral filtering algorithm, and then preprocessed for the infrared fall images after the first preprocessing according to the histogram equalization strategy to obtain the initial infrared fall images. The initial infrared fall images are labeled using an image annotation tool to obtain target infrared fall images containing the bounding rectangle of the human body. The target infrared fall images are then divided into the same dataset to obtain the target infrared fall image dataset.

3. The method of claim 2, wherein, The expression for the preset bilateral filtering algorithm is: ， ， ， ， wherein, is the pixel value of the filtered image at position x, is the pixel value of the original image at position y, is the pixel value of the original image at position x, is the support region of the filter, is the Euclidean distance between pixel points x and y, is the pixel point is the absolute value of the difference between the pixel value is the pixel value of the original image at position y, is a normalization factor, ensuring the weights sum up to 1, is a spatial weight function, indicating the influence of spatial distance between pixels, is an intensity weight function, indicating the influence of difference between pixel values, is the spatial standard deviation, is the standard deviation of intensity.

4. The method of claim 2, wherein, The expression for the histogram equalization strategy is: ， ， ， wherein, Pi is the probability density of the i-th gray level of the original image, Pi is the probability density of the i-th gray level of the original image, Ni is the number of samples in the original image with gray level i, Ni is the number of samples in the original image with gray level i, N is the total number of pixels of the image, G is the total number of gray levels, Pi is the probability density of the i-th gray level of the original image, Fi is the cumulative distribution function of the i-th gray level of the original image, with a value range of [0, 1], Pi is the probability density of the i-th gray level of the original image, G is the mapped gray level after equalization, G is the rounding integer, Gi is the i-th gray level of the original image, Gi is the i-th gray level of the original image, Gi is the i-th gray level of the original image, Ni is the number of samples in the original image with gray level i, Ni is the number of samples in the original image with gray level i.

5. The fall detection optimization method based on infrared image features according to claim 1, characterized in that, The expression for the MPDIoU loss function is: ， ， ， ， ， ， In the formula, Let MPDIoU be the loss function. The intersection-union ratio (IU) of the predicted bounding box and the ground truth bounding box is an index that combines position and shape constraints. To predict the overlap index between the bounding box and the ground truth bounding box, Let be the intersection area of ​​the two bounding boxes. Let the area be the union of the two bounding boxes. To predict the area of ​​the frame, This represents the actual area of ​​the frame. It is the square of the Euclidean distance between a diagonal vertex of the predicted bounding box and the corresponding vertex of the ground truth bounding box. It is the square of the Euclidean distance between the other diagonal vertex of the predicted bounding box and the corresponding vertex of the ground truth bounding box. , These are the coordinates of the two diagonal vertices of the prediction bounding box. , These are the coordinates of the two diagonal vertices of the real bounding box. , These are the width and height of the input image, respectively.

6. The fall detection optimization method based on infrared image features according to claim 1, characterized in that, The expression for the LAE module is: ， ， ， ， In the formula, This is the final output feature map of the LAE module. The output feature map of depthwise separable convolution. For channel attention weights, For the input feature map, This is a global average pooling operation. This is the weight matrix of the first fully connected layer. This is the weight matrix of the second fully connected layer. for Activation function for Activation function For depthwise separable convolution operations, For batch normalization operations, For activation function, This is the feature map after channel attention weighting.

7. A fall detection optimization system based on infrared image features, characterized in that, include: The acquisition module is configured to acquire infrared fall images in multiple scenarios, and to perform standardized annotation and data augmentation on each infrared fall image to obtain the target infrared fall image dataset; The module is configured to build an improved YOLOv11n network model and iteratively train and validate the improved YOLOv11n network model based on the target infrared fall image dataset to obtain a target detection model. The improved YOLOv11n network model includes the introduction of the MPDIoU loss function and the replacement of the Conv module in the improved YOLOv11n network model with a lightweight adaptive extraction LAE module. The output module is configured to input the real-time infrared fall image to be detected into the target detection model, and the target detection model outputs the fall detection result corresponding to the real-time infrared fall image.

8. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the method described in any one of claims 1 to 6.

Images