Disclosure of Invention
The following presents a simplified summary of one or more aspects in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that is presented later.
The invention provides a human lower limb gait recognition device and method. In embodiments of the present invention, the gait class is perceived by using contact force data, joint angle data, torso inclination data. Instead of dividing the gait by presetting a threshold, the technical scheme of the invention adopts a gait classifier to realize gait recognition, thereby achieving the purpose of judging the gait class.
In one aspect, the present invention provides a human lower limb gait recognition method, including: collecting joint angle data, sole pressure data and trunk inclination angle data of lower limbs of a human body in preset time; dividing the joint angle data into gait categories according to a standardized gait pattern; according to each gait class, foot sole pressure data and trunk inclination angle data corresponding to each gait class are divided; and storing a corresponding data relationship between each gait class and the plantar pressure data and the trunk inclination angle data corresponding to each gait class.
According to an embodiment of the invention, the method further comprises: acquiring foot sole pressure data and body inclination angle data of lower limbs of a human body in real time; and carrying out gait recognition on the foot sole pressure data and the trunk inclination angle data of the lower limbs of the human body collected in real time according to the corresponding data relation, and recognizing the gait class of the current time point of the human body.
According to an embodiment of the invention, the method further comprises driving the motion of the lower limbs of the human body in real time based on the gait class of the current time point of the human body.
According to an embodiment of the invention, the predetermined time is greater than one gait cycle.
According to an embodiment of the invention, the dividing the joint angle data into gait classes according to a standardized gait pattern comprises: each gait cycle in the curve of the change in joint angle over time is divided into 8 gait classes according to a standardized gait pattern.
In another aspect, the present invention provides a human lower limb gait recognition apparatus, including: the system comprises a plurality of sensors, a plurality of sensors and a control unit, wherein the sensors comprise a joint angle sensor for collecting joint angle data of lower limbs of a human body, a pressure sensor for collecting sole pressure data and an inclination angle sensor for collecting body inclination angle data; a receiving unit configured to receive data from the plurality of sensors; a processing unit configured to divide the joint angle data collected by the joint angle sensor for a predetermined time into gait classes according to a standardized gait pattern, and to divide plantar pressure data and torso inclination data collected by the pressure sensor and the inclination sensor for the predetermined time, respectively, corresponding to the gait classes according to the gait classes; and a storage unit configured to store a correspondence data relationship between each gait class and plantar pressure data and torso inclination data corresponding to the each gait class.
According to an embodiment of the invention, the processing unit is further configured to: and carrying out gait recognition on the plantar pressure data and the trunk inclination angle data of the lower limbs of the human body, which are respectively collected by the pressure sensor and the inclination angle sensor in real time, by using a gait classifier according to the corresponding data relation, and recognizing the gait class of the current time point of the human body.
According to an embodiment of the invention, the device further comprises a driving unit, wherein the processing unit is further configured to control the driving unit to realize real-time driving of the lower limb movement of the human body based on the gait class of the current time point of the human body.
According to an embodiment of the invention, the gait classifier is trained by taking as input the collected plantar pressure data and torso inclination data and as output a gait class determined from the collected joint angle data.
According to an embodiment of the invention, the gait classifier is retrained further based on the received feedback.
According to an embodiment of the invention, the predetermined time is greater than one gait cycle.
According to an embodiment of the invention, the dividing the joint angle data into gait classes according to a standardized gait pattern comprises: each gait cycle in the curve of the change in joint angle over time is divided into 8 gait classes according to a standardized gait pattern.
In still another aspect, the present invention provides an exoskeleton-assisted robot, an exoskeleton rehabilitation robot or a biped robot comprising the human lower limb gait recognition device.
As described above, by eliminating the use of the preset sensor threshold, the method and the device of the invention are more robust, and the gait subdivision capability does not depend on the number of sensors, and finally the corresponding data relationship between each gait class and the plantar pressure data and the trunk inclination angle data corresponding to each gait class is stored, thereby greatly reducing the difficulty of gait classification, being convenient for accurately judging the gait action intention of a wearer in the follow-up process, and realizing the complex gait motion follow-up.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. It is noted that the aspects described below in connection with the figures and the specific embodiments are only exemplary and should not be construed as imposing any limitation on the scope of the present invention.
Fig. 1 is a block diagram of an exemplary human lower limb gait recognition apparatus 100 according to an embodiment of the invention.
As shown in fig. 1, the human lower limb gait recognition apparatus 100 includes a sensor 102, a receiving unit 104, a processing unit 106, and a storage unit 108.
According to an embodiment of the present invention, the sensor 102 may include a plurality of sensors, such as a joint angle sensor mounted on the lower limb of the wearer, a pressure sensor attached to the sole of the wearer, and a tilt sensor mounted on the torso of the wearer, among others. The sensor 102 transmits the sensed data to the receiving unit 104, as indicated by arrow 116 in fig. 1. In one embodiment of the present invention, the sensors 102 collect joint angle data, plantar pressure data, and torso inclination data of the lower limbs of the human body over a predetermined time. For example, during the initial configuration of the human lower limb gait recognition apparatus 100, the sensors 102 collect joint angle data, plantar pressure data and torso inclination data of the corresponding human lower limbs for a predetermined time. According to one embodiment of the invention, the predetermined time is longer than one gait cycle in order to collect sufficient data. Preferably, the predetermined time comprises a plurality of gait cycles. As known to those skilled in the art, a gait cycle is the time from heel strike on one side of the body to heel strike again on the same side.
Those skilled in the art will appreciate that the sensor 102 may transmit data to the receiving unit 104 in a variety of ways. For example, the sensors 102 may transmit data via a wired or wireless connection with the receiving unit 104. Alternatively, the sensor 102 may transmit the data to another device other than the receiving unit 104, which in turn communicates with the receiving unit 104 to communicate the data. According to an embodiment of the invention, the data transmitted by the sensor 102 may be raw data or pre-processed data.
According to an embodiment of the invention, the receiving unit 104 may be various devices having the capability of receiving data from sensors, such as a wireless transceiver, a wireless receiver, various wired interfaces, and the like. According to an embodiment of the invention, the receiving unit 104, after receiving the data from the sensor 102, forwards it to the processing unit 106 for processing. According to an embodiment of the invention, the receiving unit 104 may also pre-process the sensor data in sequence before forwarding the data. Those skilled in the art will appreciate that the receiving unit 104 may forward the data to the processing unit 106 in a variety of ways.
After receiving the data from the receiving unit 104, the processing unit 106 divides the joint angle data collected by the joint angle sensor for a predetermined time into gait classes according to the standardized gait pattern, and divides the plantar pressure data and the trunk inclination angle data collected by the pressure sensor and the inclination angle sensor, respectively, for a predetermined time corresponding to the gait classes, according to the gait classes. According to an embodiment of the invention, the processing unit 106 may be various devices with processing capabilities, such as a general purpose computer, a special purpose computer, a mobile computing device such as a smart phone, and so forth.
In an embodiment of the present invention, the processing unit 106 divides the gait cycle by comparing the normalized gait patterns according to the change curve of the joint angle of the wearer. Those skilled in the art will appreciate that the standardized gait pattern can be any of a variety of prior art gait patterns, such as those described in "human walking" of Rose and j.g. gamble et al (in which a cycle of human gait is divided into 8 gait categories, such as foot strike, relative toe off, alternating anterior and posterior limbs, relative foot strike, toe off, foot convergence, tibia perpendicularity, foot strike), and the like. According to an embodiment of the present invention, the processing unit 106 divides a cycle into 8 gait classes according to the time-varying curve of the joint angle and the standardized gait pattern.
In an embodiment of the present invention, after dividing the joint angle data acquired by the joint angle sensor for the predetermined time into gait classes according to the standardized gait pattern, the processing unit 106 may divide the plantar pressure data and the trunk inclination angle data acquired by the pressure sensor and the inclination angle sensor respectively for the predetermined time corresponding to the gait classes according to the gait classes. Thus, the processing unit 106 obtains the correspondence data relationship between each gait class and the plantar pressure data and the torso inclination data corresponding to each gait class.
As described above, the human lower limb gait recognition apparatus 100 further includes the storage unit 108, and the storage unit 108 is configured to store the correspondence data relationship between each gait class and the plantar pressure data and the trunk inclination angle data corresponding to each gait class. For example, in an embodiment of the present invention, after the processing unit 106 divides the plantar pressure data and the torso inclination data corresponding to each gait class and collected by the pressure sensor and the inclination sensor respectively in the predetermined time according to each gait class, the obtained corresponding data relationship is sent to the storage unit 108 for storage and subsequent use.
According to an embodiment of the present invention, the processing unit 106 is further configured to collect the sole pressure data and the trunk inclination angle data of the lower limbs of the human body in real time, and perform gait recognition on the real-time collected sole pressure data and trunk inclination angle data of the lower limbs of the human body by using a gait classifier according to the corresponding data relationship, so as to identify the gait class of the current time point of the human body. In one embodiment of the present invention, the sensors 102 do not collect joint angle data in real time, so that during real time use of the human lower limb gait recognition device 100, the joint angle sensors can be turned off or removed from the device 100 to save power, reduce weight, and the like. However, according to an embodiment of the present invention, the joint angle sensor may also collect corresponding data in real time; the processor 106 may optionally receive, process, or not receive, joint angle data. In one embodiment, the processor 106 inputs the foot sole pressure data and the trunk inclination angle data of the lower limbs of the human body collected in real time into the gait classifier, and calculates the gait class to which the data belong according to the corresponding data relationship stored in the storage unit, so as to identify the gait class of the current time point of the human body. In an embodiment of the invention, the feature information is firstly extracted from the foot sole pressure data and the body inclination angle data of the lower limbs of the human body which are collected in real time, and then the extracted feature information is input into the gait classifier.
According to an embodiment of the invention, the gait classifier is trained by taking as input the collected plantar pressure data and torso inclination data and as output a gait class determined from the collected joint angle data. For example, in one embodiment, the processing unit 106 extracts feature information of the divided plantar pressure data and torso inclination data and uses the feature information as input data of a gait classifier to train the gait classifier.
According to yet another embodiment of the invention, the processing unit 106 may also retrain the gait classifier based on the received feedback. For example, the processing unit 106 may receive feedback from the wearer (i.e., the user) to retrain the gait classifier. For example, the wearer may provide feedback such as gait recognition that there are problems such as too fast, too slow, etc., and the processing unit 106 may retrain the gait classifier based on this feedback to re-determine the gait cycles. Subsequently, the processing unit 106 may use the retrained gait classifier for human lower limb gait recognition. Those skilled in the art will appreciate that the processing unit 106 may also retrain the gait classifier based on various other types of feedback.
According to another embodiment of the invention, in the feature extraction and the design of the gait classifier, a deep learning technology can be adopted to perform dimension reduction on the input data, and the input data with a proper dimension is selected.
As shown in fig. 1, the human lower limb gait recognition apparatus 100 can also optionally include a drive unit 110. As shown in fig. 1, the drive unit 110 is shown in a dashed box to illustrate that it is optional. In various embodiments, the processing unit 106 may control the driving unit 110 to realize real-time driving of the lower limb movement of the human body based on the gait class of the current time point of the human body. For example, in one embodiment, the drive unit 110 may be a drive unit of an exoskeleton-assisted robot or an exoskeleton rehabilitation robot. In this embodiment, the drive unit 110 may provide assistance or rehabilitation assistance to the wearer based on the gait recognized by the processing unit 106. In this way, in the actual use process, the human lower limb gait recognition device 100 can determine each state in the gait cycle according to each value measured by the sensor 102 in real time, so as to drive each joint to move, and realize real-time driving and following of the human lower limb movement. In another embodiment, the driving unit 110 may be a driving unit of a biped robot. In this embodiment, the drive unit 110 may facilitate the robot to travel based on the gait recognized by the processing unit 106.
As mentioned above, in view of the ease of implementation and accuracy of the method and apparatus of the present invention, the corresponding technical solutions are given in the above description: the gait recognition method comprises the steps of firstly dividing various gait classes according to joint angle data, further dividing and storing corresponding plantar pressure data and trunk inclination angle data according to the gait classes, and then carrying out real-time gait recognition by using the stored corresponding data relation among the gait classes, the plantar pressure data and the trunk inclination angle data. This is because the joint angle data most accurately characterizes each gait class, while the plantar pressure data and torso inclination data are more readily available in real time relative to the joint angle data.
Those skilled in the art will appreciate that this is for exemplary purposes only and that various alternatives are possible. For example, in one embodiment, only joint angle data may be collected and not plantar pressure data and torso inclination data, and as such, the correspondence between gait classes and joint angle data may be stored and joint angle data will be collected for identification at the time of subsequent real-time gait identification. In yet another embodiment, any combination of joint angle data, plantar pressure data, torso inclination data, etc. may be collected and used during gait classification or real-time gait identification. For example, joint angle data and plantar pressure data may be collected during gait classification, while only plantar pressure data is collected during real-time gait recognition; collecting joint angle data, trunk inclination angle data and plantar pressure data during gait classification, and only collecting plantar pressure data or trunk inclination angle data during gait real-time identification; only sole pressure data or trunk inclination angle data are collected during the gait classification period and the gait real-time identification period; and so on.
Fig. 2A-2B are flow diagrams of an exemplary human lower extremity gait recognition method 200, according to an embodiment of the invention. The steps of the method 200 will be described below in conjunction with the apparatus 100 shown and described in fig. 1.
As shown in FIG. 2A, the method 200 includes collecting joint angle data, plantar pressure data, and torso inclination data for a lower extremity of a human body at a predetermined time, at block 210. For example, joint angle data, plantar pressure data, and torso inclination data may be collected over a predetermined time by joint angle sensors mounted on the lower limbs of the wearer, pressure sensors attached to the soles of the feet of the wearer, and inclination sensors mounted on the torso of the wearer. In one embodiment, the predetermined time is longer than one gait cycle. Preferably, the predetermined time comprises a plurality of gait cycles.
The method 200 also includes dividing the joint angle data into gait categories according to the standardized gait pattern at block 220. For example, the processing unit 106 of fig. 1 divides the gait cycle according to the wearer's joint angle variation curve versus the standardized gait pattern. In one embodiment, one cycle is divided into 8 gait classes, as described in "human walking" by Rose and j.g. gamble et al.
The method 200 further includes, at block 230, segmenting the plantar pressure data and torso inclination data corresponding to each gait class according to the gait class. For example, after dividing the joint angle data acquired by the joint angle sensor for a predetermined time into gait classes according to the standardized gait pattern, the processing unit 106 of fig. 1 may divide the plantar pressure data and the torso inclination data acquired by the pressure sensor and the inclination sensor, respectively, for the predetermined time corresponding to the gait classes according to the gait classes. Thus, the processing unit 106 obtains the correspondence data relationship between each gait class and the plantar pressure data and the torso inclination data corresponding to each gait class.
The method 200 further includes storing, at block 240, a correspondence data relationship between each gait class and the plantar pressure data and torso inclination data corresponding to the each gait class. For example, the storage unit 108 of fig. 1 stores the resulting corresponding data relationship for subsequent use.
Optionally, the method 200 further comprises the steps 250, 260, 270, as shown in fig. 2B. Steps 250, 260, 270 are shown in dashed boxes to illustrate that they are optional.
At block 250, method 200 includes acquiring plantar pressure data and torso inclination data of the lower limbs of the human body in real time. In one embodiment of the present invention, the sensors 102 do not collect joint angle data in real time, but only plantar pressure data and torso inclination data in real time. Thus, in one example, the joint angle sensor may be turned off or removed from the device 100 to save power, reduce weight, etc., while the human lower limb gait recognition device 100 is in real-time use. However, according to an embodiment of the present invention, the joint angle sensor may also collect corresponding data in real time; the processor 106 may optionally receive, process, or not receive, joint angle data.
At block 260, the method 200 includes performing gait recognition on the real-time collected plantar pressure data and torso inclination data of the lower limbs of the human body by using a gait classifier according to the corresponding data relationship, and identifying a gait class of the current time point of the human body. In an embodiment, the processor 106 in fig. 1 inputs the foot sole pressure data and the trunk inclination angle data of the lower limbs of the human body, which are acquired in real time, into the gait classifier, and calculates the gait class to which the data belong according to the corresponding data relationship stored in the storage unit, so as to identify the gait class of the current time point of the human body. In an embodiment of the invention, the feature information is firstly extracted from the foot sole pressure data and the body inclination angle data of the lower limbs of the human body which are collected in real time, and then the extracted feature information is input into the gait classifier.
At block 270, the method 200 includes enabling real-time actuation of motion of the lower extremities of the human body based on the gait class in which the current point in time of the human body is. For example, in one embodiment, the method 200 may include providing assistance or rehabilitation assistance to the wearer via a drive unit (e.g., the drive unit 110 of fig. 1) based on a gait class of the human body at a current point in time.
In embodiments of the present invention, the gait classifier is designed based on machine learning algorithms (neural networks, support vector machines, etc.). For example, according to an embodiment of the present invention, the gait classifier is trained by taking plantar pressure data and torso inclination data as inputs and a gait class determined from joint angle change characteristics as an output. According to yet another embodiment of the invention, the gait classifier can also be retrained based on the received feedback (not shown in fig. 2A-2B). For example, the method 200 may optionally include retraining the gait classifier based on feedback received from the wearer (i.e., the user). For example, the wearer may provide feedback such as gait recognition that there are problems such as being too fast, too slow, etc. Subsequently, the method 200 may also optionally include the step of using the retrained gait classifier for human lower limb gait recognition. Those skilled in the art will appreciate that the method 200 may also retrain the gait classifier based on various other types of feedback.
Thus, the gait classification can be determined by self-learning of the algorithm by adopting a machine learning method without manually determining the threshold value of each sensor and only by taking the angle data of each joint, the pressure data of the sole and the inclination angle data of the trunk as the input of the classifier. Therefore, the method does not need to establish the relationship between the sensor state and the gait class manually and explicitly, so that the workload of manual analysis is reduced, and the algorithm has better robustness.
Those skilled in the art will appreciate that the order of operation of the above-described steps is given for illustrative purposes only and that the steps may be performed in any suitable order.
The previous description of the invention is provided to enable any person skilled in the art to make or use the invention. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the present invention is not intended to be limited to the examples and designs described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.