CN201815058U - Treadmill capable of identifying gestures - Google Patents

Abstract

The utility model discloses a treadmill capable of identifying gestures, comprising a body, a motor installed on the body, a main control system and a camera, wherein the camera is connected with the input end of the main control system, and the output end of the main control system is connected with the motor. The treadmill based on the computer vision adopts the camera as a tool for collecting the gestures of a user, the control effect identical to that of a key can be achieved when a user makes the gestures facing to the camera, and the purpose of controlling the treadmill can be achieved just through simple gestures, thereby freeing the user from the restriction of a key panel and causing operation to be convenient.

Description

Can discern the treadmill of gesture

[technical field]

The utility model relates to the treadmill field, relates in particular to a kind of treadmill that has carried based on the real-time static gesture identification interaction technique of computer vision, belongs to artificial intelligence and pattern-recognition on subject.

[background technology]

Treadmill is as a kind of indoor body-building apparatus, remedied that the outdoor sport place lacks, the incomplete shortcoming of sports facility, is people's the healthy Gospel of having brought.In modern society, treadmill almost is the indispensable sports equipment in each tame fitness center, and some booming income crowds have also had the treadmill of oneself.

Along with the increase of number of users, treadmill must provide good user's operating platform, to satisfy most of user's demand.Treadmill on the market substantially all is to allow the user by come to send corresponding order to treadmill by the button on the panel now.The user goes still also to feel light by these buttons when static, if but just on treadmill, run, rethink by them, just seem that some is painstaking, particularly when the position of position of running and key panel is far away, if the runner wants to operate treadmill, just have on the crawler belt of treadmill, run back and forth so that the finger of oneself can be by arriving button.

[utility model content]

The purpose of this utility model solves the problems of the prior art exactly, proposes a kind of treadmill of discerning gesture, can make the user cast aside the constraint of key panel, only need just can reach the purpose of controlling treadmill by several simple gestures.

For achieving the above object, the utility model proposes a kind of treadmill of discerning gesture, comprise body and the motor that is installed on the body, also comprise master control system and camera, described camera is connected to the input of master control system, and the output of master control system is connected to motor.

As preferably, described camera adopts the camera of 1.3M pixel.Recognition effect is good, the cost performance height.

The beneficial effects of the utility model: the utility model is based on computer vision, adopt the sampling instrument of camera as user's gesture, the user only need make a gesture facing to this camera, can realize the control effect identical with button, only need just can reach the purpose of control treadmill by several simple gestures, make the user cast aside the constraint of key panel, made things convenient for operation.

Feature of the present utility model and advantage will be elaborated in conjunction with the accompanying drawings by embodiment.

[description of drawings]

Fig. 1 can discern the structural representation of the treadmill of gesture for the utility model;

Fig. 2 can discern the module diagram of master control system in the treadmill of gesture for the utility model.

[specific embodiment]

As shown in Figure 1, 2, can discern the treadmill of gesture, comprise body 2, be installed in motor (not shown), master control system and camera 1 on the body 2, described camera 1 is connected to the input of master control system, and the output of master control system is connected to motor.Described camera 1 adopts the camera 1 of 1.3M pixel.Described master control system comprises real-time video input module 31, key-frame extraction module 32, skin color segmentation module 33, gesture region extraction module 34, gesture feature extraction module 35 and the gesture identification module 36 that connects successively.Described master control system adopts the embedded industrial control board of INTEL ATOM, and dominant frequency is 1.6GHz.The operating system that adopts is (SuSE) Linux OS.The distribution situation of the described skin color segmentation module 33 statistics colours of skin under the hsv color space is partitioned into area of skin color in the image according to distribution situation then.Described gesture feature extraction module 35 adopts image normalization rotary inertia (NMI) and Hu, and bending moment is not as the feature of describing gesture, and the characteristic vector of its composition is as the classification foundation of grader.Described gesture identification module 36 adopts SVMs (SVM) as grader, after using a large amount of samples that grader is carried out learning training, the result of study is saved in the file of XML form, and when discerning next time, the result that only need read out study from file gets final product.

The gesture preprocessing process has comprised key-frame extraction, image filtering, human body complexion and step such as has cut apart.Wherein image filtering has adopted median filter method.Skin color segmentation counts the distribution of the colour of skin according to the distribution histogram of human body complexion under the hsv color space, thereby has extracted the zone of gesture in image exactly, and this has done preparation for next step gesture identification.

Described gesture identification process comprises that gesture feature extracts and two steps of gesture identification.Feature extraction is meant and extracts one group of data that from gesture these group data can fully characterize gesture, are called the characteristic vector of gesture.The utility model adopted the image normalization rotary inertia (NMI) of invariancies such as having rotation, translation and Hu not bending moment gesture feature is described.

In the said method NMI be around image centroid (cx, rotary inertia cy) (functional relation between J (cx, cy)) and the picture quality m is defined as (1) formula:

$\mathrm{NMI}=\frac{\sqrt{J_{(\mathrm{cx},\mathrm{cy})}}}{m}=\frac{\sqrt{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}({(x-\mathrm{cx})}^2+{(y-\mathrm{cy})}^2)f(x,y)}}{\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}f(x,y)}---\left(1\right)$

Gesture identification is to instigate machine recognition to go out people's gesture behavior in the said method.It is divided into machine learning and two steps of classification.The utility model has adopted the instrument of SVMs (SVM) as classification.Grader is trained by the characteristic value of extracting sample at learning phase, and the relevant parameter after the preservation training.When identification, the parameter that obtains during according to training just can be divided into gesture in the predetermined class accurately, thereby realizes the identification of gesture.

Machine learning is meant how to make computer understand human behavior in the said method.It is the core of artificial intelligence, is to make computer have the fundamental way of intelligence.In the utility model, the task of machine learning is to make computer distinguish out different gestures as the people, and different gestures is divided in the inhomogeneity.

SVM is a kind of machine learning method that is come by the statistics development in the said method, is proposed by people such as Vapnik the earliest.Its core concept is: in the problem that two classes are divided, it makes the distance between two category features maximize by constructing an optimum hyperplane, thereby reaches the classification purpose; For the inseparable problem of lower dimensional space lower linear, it is transformed under the higher dimensional space by a definite Function Mapping relation, thereby makes former problem become the problem of a linear separability under higher dimensional space.

In the collected treadmill of the real-time video system that camera 1 is captured, after having carried out key-frame extraction, skin color segmentation, gesture extracted region, feature selecting and machine recognition, obtain the implication of user's gesture representative.Concrete implication is interpreted as concrete control command by system and removes to control treadmill.

Above every kind of gesture is selected the content of 200 sample gestures as machine learning respectively, the result of machine learning is saved in the XML file, and only need read out study next time from this document when discerning result just can identify gesture exactly.This benefit of bringing is to learn for every treadmill removes to select learning sample, only first learning outcome file of being preserved need be copied on the other machines to get final product.

Gesture identification accuracy rate of the present utility model is 97%, and can handle the video image of 14 frames each second, has guaranteed the real-time of identification and higher accuracy rate.Can be simply when the user moves just can control speed, the gradient of treadmill etc. by several gestures on treadmill, convenient and practical.

The foregoing description is to explanation of the present utility model, is not to qualification of the present utility model, any scheme after the utility model simple transformation is all belonged to protection domain of the present utility model.

Claims (2)

1. can discern the treadmill of gesture, comprise body and the motor that is installed on the body, it is characterized in that: also comprise master control system and camera, described camera is connected to the input of master control system, and the output of master control system is connected to motor.

2. the treadmill of discerning gesture as claimed in claim 1 is characterized in that: described camera adopts the camera of 1.3M pixel.

Images