Intelligent lifesaving wheelchair and evacuation method

Abstract

The invention belongs to the technical field of escape appliances in public places, and provides an intelligent lifesaving wheelchair and an evacuation method, and the wheelchair comprises a wheelchair main body, an audible and visual alarm device, a voice broadcast module, a gravity sensing module, a global search module, a path planning module, an automatic obstacle avoidance module, a speed control module and a wireless transmission module. The wheelchair can normally meet the moving requirements of vulnerable groups in daily conditions; when a danger occurs, whether the wheelchair takes a vulnerable individual or not can be judged, and if yes, the wheelchair directly and automatically carries people to quickly escape; if not, vulnerable individuals are searched and recognized firstly, and rapid escape is started after the vulnerable individuals are found and carried. The evacuation efficiency and success rate of vulnerable groups after accidents occur in public places can be improved.

Description

Technical Field

[0001] The invention belongs to the technical field of escape tools in public places and relates to an intelligent lifesaving wheelchair and an evacuation method. Background Art

[0002] Public places have the characteristics of highly dense population and large flow of people. Once an emergency occurs in a public place, it is easy to cause mass deaths and injuries. Vulnerable groups such as the elderly, the weak, the sick and the disabled are more likely to become victims due to their own physical conditions, self-rescue awareness and ability, social experience and other objective deficiencies. They are also the focus of public opinion. During the evacuation process, the evacuation ability of vulnerable groups will not only affect themselves, but also affect the evacuation of normal pedestrians, and thus affect the overall evacuation efficiency. Therefore, it is of great significance to pay attention to the evacuation safety of vulnerable groups in emergency situations and minimize the evacuation time.

[0003] When dealing with crises in public places, ordinary wheelchairs move slowly and require the assistance of others to control. They cannot automatically identify vulnerable people in need of help, and cannot effectively guide people with sound and light. They can meet the basic functions of daily use, but the help they provide to vulnerable people during the evacuation process is limited.

[0004] In summary, the shortcomings of existing methods include:

[0005] In an evacuation scenario, ordinary wheelchairs move slowly, require the assistance of others to control them when in use, cannot automatically identify vulnerable people who need help, and cannot effectively guide crowds with sound and light. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide an intelligent life-saving wheelchair and evacuation method to improve the evacuation efficiency and success rate of vulnerable groups after accidents in public places. The wheelchair can normally meet the mobility needs of vulnerable groups under daily circumstances; when danger occurs, it can determine whether the wheelchair has already taken a vulnerable individual, and if so, it will directly and automatically carry the person for rapid escape; if not, it will first search and identify the vulnerable individual, and then start rapid escape after finding and carrying the vulnerable individual.

[0007] The technical solution adopted by the present invention is as follows:

[0008] An intelligent life-saving wheelchair, comprising:

[0009] A wheelchair body, and an audible and visual alarm device, a voice broadcast module, a gravity sensor module, a global search module, a path planning module, an automatic obstacle avoidance module, a speed control module and a wireless transmission module mounted on the wheelchair body; the wheelchair includes a daily mode and a life-saving mode. In the daily mode, the wheelchair is used to meet the daily mobility needs of vulnerable groups. When the public safety alarm device is activated, the wheelchair is switched from the daily mode to the life-saving mode to carry the person in need of rescue for rapid evacuation; wherein:

[0010] The sound and light alarm device transmits data with the safety alarm device in the public place through the wireless transmission module;

[0011] The voice broadcast module is used to broadcast evacuation information;

[0012] The gravity sensor module is used to detect the gravity signal transmitted from the seat to determine whether the chair is occupied;

[0013] The global search module transmits data with the video surveillance center of the public place through the wireless transmission module to determine the spatial location of the person in need of rescue;

[0014] The path planning module is used to receive signals from the gravity sensing module and the global search module, and plan the optimal path to the safe exit or the optimal path to the location of the person in need of rescue;

[0015] The automatic obstacle avoidance module is used to receive signals from the path planning module and avoid obstacles along the way;

[0016] The speed control module is used to automatically control the speed of the wheelchair body during travel;

[0017] The wireless transmission module is used for data transmission between the wheelchair and a safety alarm device and a video monitoring center in a public place.

[0018] Furthermore, when the wheelchair is switched from daily mode to life-saving mode, the sound and light alarm device is activated synchronously and starts flashing to alarm.

[0019] Furthermore, the voice broadcast module plays key evacuation information in a loop.

[0020] Furthermore, the global search module constructs a global personnel movement model based on video surveillance information in public places through a computer vision core algorithm, and analyzes and determines the people in need of rescue and their locations.

[0021] Global search algorithm: target detection YOLOv5 + target tracking DeepSORT + spatial coordinate conversion

[0022] Furthermore, the path planning module receives the signal from the gravity sensor module, and when the wheelchair is occupied, the path planning module automatically plans the optimal path to the safe exit; when the wheelchair is not occupied, the path planning module receives the signal transmitted by the global search module, and automatically plans the optimal path to the location of the person in need of rescue. When the wheelchair successfully carries the person in need of rescue, the optimal path to the safe exit is re-planned.

[0023] Furthermore, the speed control module can flexibly adjust the running speed of the wheelchair according to actual conditions to minimize the overall movement time of the wheelchair.

[0024] Furthermore, the wireless transmission module can realize real-time data transmission and processing.

[0025] An evacuation method using the above-mentioned intelligent life-saving wheelchair comprises the following steps:

[0026] After the wheelchair receives the safety alarm signal, it automatically switches to life-saving mode;

[0027] The wheelchair uses the gravity sensor module to determine whether someone is already sitting in it;

[0028] If the wheelchair is not occupied, the global search module will be used to find the person in need of assistance and automatically go to the location of the person in need of assistance;

[0029] When the wheelchair successfully carries the person in need of rescue, the optimal path is planned through the path planning module, and evacuation is carried out through the automatic obstacle avoidance module and speed control module.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] For wheelchairs placed in public places, the present invention not only includes all the functions of ordinary wheelchairs, but also takes into account the function of helping vulnerable groups to evacuate in emergency scenarios, thereby improving the evacuation efficiency and success rate of vulnerable groups, and further improving the evacuation efficiency of normal people.

[0032] Ordinary wheelchairs require assistance from others to control, but the present invention does not require human assistance during the evacuation process and can achieve fully automatic operation.

[0033] In view of the slow running speed of ordinary wheelchairs, the present invention takes into account a speed control module to flexibly adjust the running speed according to actual conditions during the movement.

[0034] As ordinary wheelchairs cannot automatically search for vulnerable groups, the present invention can automatically find those in need of rescue and assist them in completing the evacuation smoothly.

[0035] As ordinary wheelchairs do not take into account the sound and light alarm and guidance issues, the present invention is equipped with a sound and light alarm and voice broadcast device to remind and comfort people along the way to evacuate efficiently, which can reduce people's reaction time and alleviate people's panic. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 The figure is a schematic diagram of an intelligent life-saving wheelchair according to an embodiment of the present invention.

[0037] Figure 2 This is a flow chart of the target detection YOLOv5 algorithm in an embodiment of the present invention.

[0038] Figure 3 4 is a structural diagram of the CBS module within the feature extraction layer in an embodiment of the present invention.

[0039] Figure 4 Module structure diagram of module C3 inside the feature extraction layer in an embodiment of the present invention.

[0040] Figure 5 It is a structural diagram of the SPPF module in an embodiment of the present invention.

[0041] Figure 6 Schematic diagram of the actual position of the target in an embodiment of the present invention.

[0042] Figure 7 Flow chart of the DeepSORT algorithm for target tracking in an embodiment of the present invention.

[0043] Figure 8 The figure is a flow chart of extracting and calculating motion information in cascade matching in an embodiment of the present invention.

[0044] Fig. 9 4 is a flowchart of extracting and calculating appearance information in cascade matching in an embodiment of the present invention.

[0045] Fig.10 This is an overall flow chart of cascade matching in an embodiment of the present invention.

[0046] In the figure, 1. sound and light alarm device, 2. wireless transmission module, 3. gravity sensor module, 4. wheelchair body, 5. path planning module, 6. automatic obstacle avoidance module, 7. voice broadcast module, 8. global search module, 9. speed control module. DETAILED DESCRIPTION

[0047] The principles and features of the present invention are described below in conjunction with the accompanying drawings and specific embodiments. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0048] The present invention provides an intelligent life-saving wheelchair to improve the evacuation efficiency and success rate of vulnerable groups after accidents in public places. The wheelchair can normally meet the mobility needs of vulnerable groups under normal circumstances; when danger occurs, it can determine whether the wheelchair has already been occupied by a person in need of rescue, and if so, it will directly and automatically carry the person for rapid escape; if not, it will first search and identify the person in need of rescue, and then start rapid escape after finding and carrying the person.

[0049] Example 1

[0050] like Figure 1As shown, the present invention provides an intelligent life-saving wheelchair, comprising: a wheelchair body 4, an acoustic and optical alarm device 1, a voice broadcast module 7, a gravity sensor module 3, a global search module 8, a path planning module 5, an automatic obstacle avoidance module 6, a speed control module 9 and a wireless transmission module 2. The wheelchair includes a daily mode and a life-saving mode. In the daily mode, it is used to meet the daily mobility needs of vulnerable groups; in the life-saving mode, it is used to carry people in need of rescue for rapid evacuation.

[0051] The wheelchair body 4 is composed of a seat surface, a backrest surface and wheels. In order to meet the needs of safe evacuation, the wheelchair body 4 is equipped with an audible and visual alarm device, a voice broadcast module, a gravity sensor module, a global search module, a path planning module, an automatic obstacle avoidance module, a speed control module and a wireless transmission module, each of which is equipped with a processor.

[0052] The sound and light alarm device is linked to the safety alarm device in the public place through a wireless transmission module. When the safety alarm device in the public place is activated, the wheelchair switches from daily mode to life-saving mode, and the sound and light alarm device of the wheelchair is activated synchronously; the sound and light alarm device starts to flash and alarm, and links to the voice broadcast module. The voice uses a friendly female voice to play key information such as what happened in the place and which direction to evacuate in a loop from small to large.

[0053] The gravity sensing module is used to detect the gravity signal transmitted by the gravity sensor on the seat to determine whether the chair is occupied.

[0054] The global search module is connected to the video surveillance center of public places through the wireless transmission module to obtain all video information in public places. The core computer vision algorithm then identifies vulnerable groups such as the elderly, the weak, the sick, the disabled, and pregnant women, and builds a global personnel motion model to further calculate the movement speed of these people. After analysis, the spatial positions of the elderly, the weak, the sick, the disabled, and pregnant women with limited mobility who are slower are determined.

[0055] The computer vision core algorithm described in the present invention can adopt any target detection and target tracking algorithm in the prior art that can realize the function of the present invention. A specific example is provided below: including target detection YOLOv5, target tracking DeepSORT and space coordinate conversion algorithm.

[0056] Object Detection YOLOv5:

[0057] YOLOv5 is a single-stage target detection algorithm based on convolutional neural networks. Its network structure mainly includes input layer, feature extraction layer, feature fusion layer and prediction output layer. The function of the input layer is to preprocess the input data, and use a variety of data enhancement methods to improve the diversity of data during training, improve the generalization ability of the network after training, and prevent overfitting; the feature extraction layer uses a unique network structure to extract the features of the enhanced data for use in subsequent reasoning processes. The input data outputs feature maps of different scales at each stage of the feature extraction network and is fused in the feature fusion layer. The fused features of different scales are input into the three prediction output layers for target category and coordinate box position prediction. The YOLOv5 algorithm flow chart is shown below. Figure 2 shown.

[0058] The feature extraction network structure in YOLOv5 is as follows Figure 1 As shown in the Backbone section, the CBS module is the most basic module of the network, which consists of a convolutional layer, a normalization layer (BN) and a SiLU activation function. Its structure is as follows Figure 3 The C3 module is an improved module based on the CSPNet concept. Its structure is as follows Figure 4 As shown in Figure 1, the structure is composed of three standard convolutional layers and N Bottleneck modules in a residual connection. The use of C3 modules can effectively reduce the amount of calculation while ensuring the network feature extraction capability, and reduce repeated gradient information during the network back propagation process.

[0059] The SPPF structure is as follows Figure 5 As shown in the figure, the input feature map passes through three maximum pooling layers with the same convolution kernel size in sequence, and the results after each pooling are concatenated. Experiments have shown that this type of pyramid structure can increase the receptive field of the output feature map, obtain features of more scales, and has a faster calculation speed than the SPP structure.

[0060] After the input image enters the network, a CBS layer with a convolution kernel size of 6*6 is first used to reduce the image size and expand the channel dimension, effectively reducing the number of parameters that need to be calculated. The feature map is then downsampled multiple times through multiple CBS modules and C3 modules in the network and its features are extracted. Finally, an SPPF module is used to obtain an adaptive size feature map with stronger feature expression capabilities.

[0061] The feature fusion network structure in YOLOv5 is as follows Figure 2 As shown in the Neck part, the upsampling operation of the output feature layer adopts the nearest neighbor interpolation method.

[0062] The existence of the feature fusion layer provides the network with the ability to obtain multi-scale features. Theoretically, as the depth of the convolutional neural network increases and the number of downsampling times increases, the feature map will obtain a larger receptive field and features that carry more semantic information. However, due to the low resolution, detailed information cannot be obtained, so it can only be used to detect larger targets in target detection; while the shallow network often obtains some features with small receptive fields but rich detailed information, which can be used to detect smaller targets. In order to make full use of the advantages and disadvantages of deep and shallow neural networks to detect targets of different scales, the Neck layer continuously upsamples the feature map obtained by the feature extraction network and splices it with the feature map of the corresponding size in the Backbone, and continues to use the C3 module to extract features from the spliced ​​new feature map, so as to obtain a feature map with higher quality and stronger semantic information. Finally, a top-down fusion method is used to make the features obtain stronger positioning capabilities, and finally the feature maps of different scales are input into the prediction output layer for classification and positioning.

[0063] The prediction output layer in YOLOv5 is as follows Figure 1 As shown in the Head part, the prediction output layer receives feature maps of different sizes output by the previous network and uses these feature maps to predict large, medium, and small objects respectively.

[0064] After each feature map is input into the prediction output layer, a convolution kernel of size 1*1 is used to convert the feature map into (N+4+1)*3 channels, where N represents the number of predictable categories and 4 represents the location information of the category (b x ,b y ,b w ,b h ), 1 indicates the confidence that the target belongs to this category, and 3 indicates 3 anchors of different sizes used to predict the target.

[0065] b x =(2*σ(t x )-0.5)+c x (1)

[0066] b y =(2*σ(t y )-0.5)+c y (2)

[0067] The location information of each category is b x , b y The calculation formula is shown in formula (1) and (2), where c x , c y Represents the coordinates of the upper left corner of the Grid grid, t x and t yRepresents the offset of the predicted target from the upper left corner of the Grid. To prevent the predicted position offset from being too large, the Sigmoid function is first used to limit its size to (0, 1).

[0068] like Figure 6 As shown in (a), when the center point of the real target happens to fall at the extreme coordinate position of the upper left corner or lower right corner of the grid, according to the characteristics of the Sigmoid function, the network prediction value needs to be positive infinity or negative infinity to be accurately predicted, but the general network will not output such extreme values, and when the output value is close to 0 or close to 1, it is easy to cause the gradient to disappear during the back propagation process. At this time, the predicted output value can be multiplied by 2 and then subtracted by 0.5, such as Figure 6 As shown in the yellow box in (b), by controlling the output range to (0.5, 1.5), the modified network offset can easily take extreme values ​​such as 0 or 1.

[0069] b w =p w *(2*σ(t w )) 2 (3)

[0070] b h =p h *(2*σ(t h )) 2 (4)

[0071] In order to prevent problems such as gradient explosion and unstable training, the width and height of the predicted target in the position information are also restricted. The formulas are shown in (3) and (4). The Sigmoid function is used to limit its range to (0, 1) and a series of operations are performed to obtain the multiplication factor. The multiplication factor is multiplied by the corresponding anchor width and height to obtain the final predicted width and height.

[0072] Object Tracking DeepSORT:

[0073] The algorithm flow of DeepSORT is as follows Figure 7 As shown in the figure, the video sequence is detected by the detector, and then the target box information is sent to the DeepSORT tracker. In the first frame of the video, a Kalman filter is used to initialize a new track for all detections. At this time, the track state is unconfirmed. When the track matches the detection successfully for three consecutive frames, it will be confirmed. At the t-1 frame, the Kalman filter is used to predict the position of all tracks in the t-1 frame according to the track state. The predicted track is recorded as Tracker t-1 , where the track of the confirmation state is denoted as ConTracker t-1 , the unconfirmed track is recorded as UnConTracker t-1When the detection box information of the tth frame is input into the tracker, it is recorded as Detection t , using ConTracker t-1 Detection t The appearance information and motion information of the object are cascade matched. The cascade matching outputs three sets: successfully matched detections and tracks, unmatched tracks, and unmatched detections. In order to further improve the tracking accuracy, unmatched tracks, unmatched detections, and UnConTracker are used to t-1 The motion information of the two images is then matched again using IOU, and three sets are finally output: successfully matched detections and trajectories, unmatched trajectories, and unmatched detections.

[0074] The detections and trajectories that are successfully matched in the two matching processes are input into the Kalman filter, the trajectory is updated using the detection status information, and the updated trajectory is added to the trajectory set Tracks at time t t For unmatched detections, initialize a new non-confirmed track and add it to Tracks t For unmatched tracks, if it is UnConTracker t-1 Or a ConTracker that has exceeded its maximum lifespan threshold t-1 , then delete it, if the ConTracker does not exceed the maximum life threshold t-1 , then keep it and add it to Tracks t When Tracks t After all members have joined, the Kalman filter is used to predict their positions, and the prediction result is recorded as Tracker t .

[0075] Cascade matching is the core step of the DeepSORT algorithm, which effectively solves the problem of divergence of Kalman filter prediction results caused by long-term occlusion of the target.

[0076] Cascade matching uses the appearance information and motion information of the target to match new detections and trajectories respectively.

[0077] 1) Motion information: The extraction and calculation process of motion information in cascade matching is as follows Figure 8 As shown, the Mahalanobis distance between the detection boxes of all targets at time t and the prediction boxes of all trajectories at time t-1 is calculated. The calculation formula is shown in formula (5):

[0078]

[0079] Where i represents the i-th track and j represents the j-th detection. i and S iThey represent the predicted position and predicted covariance matrix of the trajectory at time t-1 respectively. j Indicates the detection box information detected at time t. (1) The smaller the value of (i, j), the higher the similarity of motion features between the detection and the trajectory, and the greater the correlation match.

[0080] 2) Appearance information: The extraction and calculation process of appearance information in cascade matching is as follows Fig. 9 As shown in Figure 1, when the detection images of all targets at time t are input into the tracker, the ReID network is used to extract features of each target, and the obtained feature matrix is ​​recorded as M. At the same time, the feature vectors of all trajectories are extracted to form the feature matrix N. The cosine similarity between M and N is calculated as shown in formula (6):

[0081]

[0082] where r j represents the feature vector of the jth detection, represents all feature vectors of the i-th detection. (2) The smaller the value of (i, j), the higher the similarity of appearance features between the detection and the trajectory, and the greater the correlation match.

[0083] In the entire cascade matching process, the cost matrix calculation formula of the association problem is shown in formula (7):

[0084] c i，j =λd (1) (i, j)+(1-λ)d (2) (i,j) (7)

[0085] After obtaining the cost matrix, the Hungarian algorithm is used to match the trajectories in the order of trajectory update time from recent to far. The overall process of cascade matching is as follows: Fig.10 shown.

[0086] 3) Kalman filter

[0087] Kalman filter plays a very important role in tracking algorithms. In the DeepSORT algorithm, Kalman filter is mainly responsible for two tasks: position prediction and state update of trajectory. In the first frame of the tracking process, Kalman filter will use the coordinate information of the detection box as a direct measurement of the object state and initialize a trajectory for all detections. Each trajectory is defined as an eight-dimensional state space x = (c x , c y , k, h, v x , v y , v k , v h ), where c x , cy is the center coordinate of the detection box, k is the aspect ratio of the detection box, h is the height of the detection box, and v x , v y , v k , v h Respectively represent the speed change values ​​of the above four values.

[0088] a) Trajectory Position Prediction

[0089] At time t all Tracks t After all members have joined, the Kalman filter is used to predict their positions and obtain Tracks t The possible positions of all members at time t+1 are recorded as Tracker t The process of Kalman filter predicting the position of the trajectory is shown in formulas (8) and (9):

[0090] x′ t =Fx t (8)

[0091] P′ t =FP t F T +Q (9)

[0092] In formula (8), x t and x′ t They represent the state information of the trajectory at time t (c x , c y , k, h, v x , v y , v k , v h ) and the state information of the trajectory after prediction; F is the state transfer matrix, which is used to infer the state of the next moment from the previous moment. The uniform motion model is used in DeepSORT, so the formula of the state transfer matrix is ​​shown in formula (10), d t Indicates the displacement between the previous and next frames.

[0093]

[0094] In formula (9), P t and P′ t The covariance matrices before and after the prediction are used to represent the uncertainty of the target position. The larger the number of the covariance matrix, the higher the uncertainty of the position. t The uncertainty P in the previous frame t It is calculated together with the system noise Q. When the system is initialized, the noise matrix Q is usually assigned a very small value.

[0095] After obtaining the mean and variance of the target predicted by the Kalman filter for the next frame, they can be used to match new detections in the next frame.

[0096] b) Track status update

[0097] At time t, the detection information of all targets, after cascade matching and IOU matching with the trajectory, will be input into the Kalman filter, and the Kalman filter will be used to update the status of the trajectory according to the detection information.

[0098] D m =d-Hx′ t-1 (11)

[0099] D c =HP′ t-1 H T +R (12)

[0100]

[0101] The mean error D between detection and trajectory within the matching item is calculated using formula (11): m First, the measurement matrix H is used to transform the mean vector x′ of the trajectory prediction at the previous moment t-1 Map to the detection space, and then calculate the detected position information d = (dc x , dc y , dk, dh) and the difference between the trajectory.

[0102] Use formula (12) to convert the covariance matrix P′ predicted at the previous moment t Mapped to the detection space, and add a detector noise matrix R initialized with arbitrary values. This results in a covariance matrix D containing the detection noise c .

[0103] When the mean error D is obtained m And the covariance matrix D of the detection c After that, the Kalman filter can start updating the trajectory state.

[0104]

[0105] First, the Kalman gain K is calculated using formula (14). Its main function is to use the covariance matrix P′ of the trajectory prediction at the previous moment t-1 The covariance matrix D of the detection c To weigh the importance of prediction trajectory and detection in the updating process.

[0106] After calculating the Kalman gain, the trajectory state update process is completed using formulas (15) and (16).

[0107] x t = x′ t-1 +KD m (15)

[0108] P t =(I-KH)P′ t-1 (16)

[0109] The Kalman gain is used in formula (15) to calculate the mean error D m Correction is performed and the state prediction value of the trajectory at time t-1 is added to obtain the state information x of the trajectory at time t t Formula (16) uses the Kalman gain to update the covariance matrix of the trajectory to obtain P t . After trajectory update, x is obtained t and P t The Kalman filter can be used to predict the trajectory position.

[0110] Space coordinate conversion: Use linear transformation method to convert video coordinates into real space coordinates.

[0111] x and y are video coordinates, and X and Y are real coordinates. The relationship between the two satisfies the following mapping relationship:

[0112]

[0113] Among them, L1~L8 are conversion coefficients.

[0114] By collecting and labeling a dataset of images of vulnerable groups such as the elderly, the weak, the sick, the disabled, and pregnant women, the dataset needs to include these targets from various angles and scenes. The YOLOv5 model is trained using this dataset to enable it to accurately identify targets such as the elderly, the weak, the sick, the disabled, and pregnant women.

[0115] The YOLOv5 detection results are used as the input of DeepSORT to achieve real-time tracking. Whenever YOLOv5 detects a target, DeepSORT tracks the position of each target in the video by assigning an ID. The ReID network is used to enhance the target's re-identification ability, ensuring that the target can be accurately tracked even when it is blocked or partially disappeared. The moving speed of the target is calculated by tracking, and the current target is further confirmed as a vulnerable group based on the moving speed, preventing the YOLOv5 algorithm from identifying errors.

[0116] Combining YOLOv5 and DeepSORT, it is possible to identify and locate vulnerable groups in videos. And through the spatial coordinate conversion method, the location information of vulnerable groups in videos can be converted into real-world coordinates, which facilitates the path planning of smart wheelchairs.

[0117] The path planning module receives the signal transmitted by the gravity sensor module. When the gravity sensor module determines that the chair is occupied, it plans the optimal path to the safe exit and automatically drives the wheels to go there. When the gravity sensor module determines that the chair is not occupied, it receives the signal transmitted by the global search module, obtains the spatial position of the nearest person in need of rescue, plans the optimal path to the location of the person in need of rescue, automatically drives the wheels to provide help, and after successfully carrying the target individual, re-plans the optimal path to the safe exit and automatically drives the wheels to go there. The algorithm used by the path planning module is A * Algorithm, A * The algorithm is an algorithm for finding the shortest path on a graph, which combines heuristics, dynamic programming, and the shortest path algorithm in graph theory. * The algorithm combines the advantages of determinism and heuristics, exploring possible solutions while using estimated information to avoid many useless solutions.

[0118] A * The algorithm uses the following function to calculate the priority of each node in the space. The smaller f(n) is, the greater the priority of node n being selected, f(n) = g(n) + h(n).

[0119] Among them, f(n) is the total moving cost from the starting point to the target point; g(n) is the moving cost from the starting point to the current point; h(n) is the moving cost from the current point to the target point.

[0120] The automatic obstacle avoidance module receives the signal from the path planning module, and follows the path set by the path planning module to avoid crowds and other obstacles that may appear along the way in time to ensure the safe travel of the wheelchair.

[0121] The speed control module controls the speed of the wheelchair body during travel, and flexibly adjusts the running speed according to the actual situation during the movement, so as to minimize the overall movement time of the wheelchair and improve the evacuation efficiency while ensuring safety. The wireless transmission module is used for real-time data transmission and processing between the wheelchair and the safety alarm device and video monitoring center in public places.

[0122] The functions of each module of the present invention are not limited to a specific algorithm implementation, but can flexibly adopt including but not limited to traditional machine learning algorithms, deep learning algorithms, and any other algorithms that can meet the performance requirements of the present invention. Through this design, the present invention can adapt to different application scenarios and requirements, providing a high degree of flexibility and scalability.

[0123] Example 2

[0124] An evacuation method using the above-mentioned intelligent life-saving wheelchair comprises:

[0125] The wheelchair of the present invention is similar in appearance to a common wheelchair and is usually placed in a special wheelchair area in a public place. It can be used by special people as a common wheelchair in public places for free, but it cannot exceed a certain range of the special wheelchair area. When an emergency occurs, the wheelchair is activated by the alarm device in the public place and the wheelchair switches to life-saving mode. The sound and light alarm device starts flashing and alarms, and the voice broadcast module is linked. The voice uses a friendly female voice to play relevant safety information and life-saving instructions in a loop from small to large, reminding people to evacuate safely.

[0126] At the same time, the wheelchair gravity sensor module senses whether someone is using the wheelchair. If someone is using it, it will combine the path planning module, automatic obstacle avoidance module and speed adjustment module to lead the person to automatically drive the wheels to a safe area; if no one is using it, it will combine the global search module, path planning module, automatic obstacle avoidance module and speed adjustment module to automatically drive the wheels to the nearest vulnerable individual to provide assistance. After successfully carrying the target individual, it will automatically drive the wheels to a safe area. After the target person is delivered to a safe area, the smart lifesaving wheelchair returns to the scene to continue helping the vulnerable group until all vulnerable people are evacuated.

[0127] When the scene returns to normal, the lifesaving wheelchair automatically navigates back to the specific location in the wheelchair-only area to wait for the next use.

[0128] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An intelligent life-saving wheelchair, characterized in that: include: A wheelchair body, and an audible and visual alarm device, a voice broadcast module, a gravity sensor module, a global search module, a path planning module, an automatic obstacle avoidance module, a speed control module and a wireless transmission module mounted on the wheelchair body; the wheelchair includes a daily mode and a life-saving mode. In the daily mode, the wheelchair is used to meet the daily mobility needs of vulnerable groups. When the public safety alarm device is activated, the wheelchair is switched from the daily mode to the life-saving mode to carry the person in need of rescue for rapid evacuation; wherein: The sound and light alarm device transmits data with the safety alarm device in the public place through the wireless transmission module; The voice broadcast module is used to broadcast evacuation information; The gravity sensor module is used to detect the gravity signal transmitted from the seat to determine whether the chair is occupied; The global search module transmits data with the video surveillance center of the public place through the wireless transmission module to determine the spatial location of the person in need of rescue; The path planning module is used to receive signals from the gravity sensing module and the global search module, and plan the optimal path to the safe exit or the optimal path to the location of the person in need of rescue; The automatic obstacle avoidance module is used to receive signals from the path planning module and avoid obstacles along the way; The speed control module is used to automatically control the speed of the wheelchair body during travel; The wireless transmission module is used for data transmission between the wheelchair and a safety alarm device and a video monitoring center in a public place.

2. The intelligent lifesaving wheelchair according to claim 1, characterized in that: When the wheelchair is switched from the daily mode to the life-saving mode, the sound and light alarm device is activated synchronously and starts flashing to alarm.

3. The intelligent lifesaving wheelchair according to claim 1, characterized in that: The voice broadcast module plays key evacuation information in a loop.

4. The intelligent lifesaving wheelchair according to claim 1, characterized in that: The global search module constructs a global personnel movement model based on video surveillance information in public places through a computer vision core algorithm, and analyzes and determines the people in need of rescue and their locations.

5. The intelligent lifesaving wheelchair according to claim 1, characterized in that: The path planning module receives the signal from the gravity sensor module. When the wheelchair is occupied, the path planning module automatically plans the optimal path to the safe exit. When the wheelchair is not occupied, the path planning module receives the signal transmitted by the global search module, automatically plans the optimal path to the location of the person in need of rescue, and when the wheelchair successfully carries the person in need of rescue, re-plans the optimal path to the safe exit.

6. The intelligent lifesaving wheelchair according to claim 1, characterized in that: The speed control module can flexibly adjust the running speed of the wheelchair according to actual conditions, thereby reducing the overall movement time of the wheelchair as much as possible.

7. The intelligent lifesaving wheelchair according to claim 1, characterized in that: The wireless transmission module can realize real-time data transmission and processing.

8. An evacuation method using the intelligent lifesaving wheelchair according to any one of claims 1 to 7, characterized in that: The following steps are involved: After the wheelchair receives the safety alarm signal, it automatically switches to life-saving mode; The wheelchair uses the gravity sensor module to determine whether someone is already sitting in it; If the wheelchair is not occupied, the global search module will be used to find the person in need of assistance and automatically go to the location of the person in need of assistance; When the wheelchair successfully carries the person in need of rescue, the optimal path is planned through the path planning module, and evacuation is carried out through the automatic obstacle avoidance module and speed control module.

9. The method according to claim 8, characterized in that During the wheelchair evacuation process, the sound and light alarm device and voice broadcast module provide evacuation information broadcast and guidance.

Images