CN103679130B - Hand method for tracing, hand tracing equipment and gesture recognition system - Google Patents

Abstract

The present invention discloses a kind of hand method for tracing, hand tracing equipment and gesture recognition system.Hand method for tracing includes：Judge whether covered in the frame before present frame；According to masking judging result, proficiency target is selected in the hand target adaptively tracked from frame above as template；And template matching method is carried out to obtain the hand target of present frame for present frame based on the template.

Description

Hand tracking method, hand tracking device and gesture recognition system

technical field

The present invention relates to the field of image recognition, and more particularly to a hand tracking method, a hand tracking device and a gesture recognition system.

Background technique

As computers and numerous portable smart devices become more and more indispensable in people's lives today, people will desire more natural and efficient interactions between humans and computers. Therefore, a gesture remote control system as a free-touch human-computer interaction interface will become an attractive choice.

Basically, the gesture remote control system will track the hands and analyze meaningful hand expressions, if they are recognized as one of the pre-defined gestures, the corresponding operation command will be triggered to run a certain application. It can be seen that a robust hand tracking method as a premise is very important for analyzing hand movements and recognizing gestures. In addition, for the implementation of low-power devices, the complexity of the tracking algorithm also needs to be considered.

Currently, certain algorithms or their variants have been introduced for real-time hand tracking. For example, the Optical Flow algorithm reflects frame changes by appending to the velocity vectors of moving pixels in the image, however, this requires a lot of computation; the CamShift algorithm can effectively track the probability distribution of dynamic changes in the visual scene, however, this Rely on proper brightness and will not be robust when there are other objects with similar color to skin color during tracking; Template Matching algorithm can give However, in some cases, the tracking accuracy is not satisfactory.

Therefore, there is a need for a hand tracking method capable of high tracking accuracy even when the hand crosses over some object having a color similar to a skin color, and a gesture remote control system applying the hand tracking method.

Contents of the invention

According to one aspect of the present invention, a hand tracking method is provided, including: judging whether occlusion occurs in the frame before the current frame; adaptively selecting a hand from the hand targets tracked in the previous frame according to the occlusion judgment result The target is used as a template; and a template matching method is performed on the current frame based on the template to obtain the hand target of the current frame.

In one embodiment, adaptively selecting the hand target as a template includes: if occlusion occurs in a frame before the current frame, the hand target obtained in the frame immediately before the occlusion occurs is used as the template; if the current frame If no occlusion occurs in the previous frame, the hand object obtained in the previous frame of the current frame is used as the template.

In one embodiment, the template matching method includes: defining a search range and a plurality of candidate hand ranges in the skin color image of the current frame; The candidate hand ranges that best match the template are used as the hand targets tracked in the current frame.

In one embodiment, defining the search range includes: defining an initial search range by expanding the periphery of the hand target tracked in the previous frame by a predetermined number of pixels in the skin color image of the current frame; and The initial search range is shrunk to the minimum rectangle containing the hand target tracked in the previous frame and the current skin color difference image representing the motion difference between the skin color image of the current frame and the skin color image of the previous frame as the search scope.

Wherein, the template matching method further includes: defining a plurality of first candidate hand ranges in the search range, these first candidate hand ranges have the same size as the template, and, between the skin color image representing the current frame and the previous A second candidate hand range is defined in the current skin color difference image of the motion difference between the skin color images of a frame, and the second candidate hand range has the same size as the template, and is executed cyclically for the plurality of first candidate hand ranges Matching and judging processing until the multiple first candidate hand ranges have been subjected to matching and judging processing, thereby determining the candidate hand range that best matches the template as the hand target tracked in the current frame. The matching and judging processing includes: Calculating the average value of the absolute difference between the first candidate hand range and the value of each pixel of the template as the first error; if the first error is greater than the first predetermined threshold, the current first candidate hand range is judged as not matching the template and thereby being excluded; if the first error is less than the first predetermined threshold, calculating a second error by subtracting the first error from the second candidate The difference obtained by multiplying the average value of each pixel value of the hand range by the value obtained by the correction factor; if the second error is smaller than the second predetermined threshold, the current first candidate hand range is determined to match the template , and the value of the second error and the average value of each pixel value of the current first candidate hand range are respectively used as the second predetermined threshold and the second candidate hand value in the matching judgment process for the next first candidate hand range. The average of the values for each pixel of the range.

In one embodiment, in the case that occlusion does not occur in the frame preceding the current frame, the correction factor is a first predetermined value, and in the case that occlusion occurs in the frame preceding the current frame, the correction factor is the first predetermined value. The sum of the two predetermined values and the percentage of the average value of each pixel value in the hand object obtained from the skin color difference image of the frame immediately before the occlusion occurs and the previous frame.

In one embodiment, the hand tracking method further includes: judging whether occlusion occurs in the current frame, and the judgment includes: if no hand target is tracked in the current frame, directly determining that occlusion does not occur in the current frame; If the hand target is tracked in the frame, the occlusion condition judgment is performed to determine whether occlusion occurs in the current frame.

In one embodiment, the occlusion condition judgment includes judging whether the following conditions are satisfied: (((PreOccDiff<=A)&&(DeltaDiff>B))||((PreOccDiff>A)&&(DeltaDiff>C)))&& (OccNum<D), wherein, PreOccDiff represents the average value of the value of each pixel in the hand target obtained in the frame before the occlusion occurs and the skin color difference image of the previous frame, DeltaDiff is the difference between PreOccDiff and AvgTarDiff, and AvgTarDiff is The average value of the value of each pixel in the current skin color difference image and the candidate hand range that matches the template most, OccNum represents the number of times that continuous occlusion occurs in front of the current frame, wherein, C is greater than B,

And wherein, if the above condition is satisfied, it is determined that occlusion has occurred in the current frame, and if the above condition is not satisfied, it is determined that occlusion has not occurred in the current frame.

According to the second aspect of the present invention, there is provided a hand tracking system, which includes: judging means for judging whether occlusion occurs in the frame before the current frame; Selecting a hand object from the tracked hand objects as a template; and obtaining means for performing a template matching method on the current frame based on the template to obtain the hand object of the current frame.

According to a third aspect of the present invention, there is provided a gesture recognition system, the gesture recognition system comprising the above-mentioned hand tracking device.

According to the hand tracking method and the hand tracking device of the present invention, it is possible to have high tracking accuracy even when the hand crosses over some object having a color similar to the skin color, thereby realizing preparatory gesture recognition.

Description of drawings

Fig. 1 shows the frame diagram of general gesture recognition system;

Fig. 2 shows an illustration of a tracking error occurring when a hand moves in front of a face in a traditional template matching algorithm;

Fig. 3 shows the flowchart of the hand tracking method according to the present invention;

Figure 4 shows a more detailed flowchart of the hand tracking method according to the present invention; and

FIG. 5 shows an illustration of the tracking result of the hand tracking method when the hand moves in front of the face according to the present invention.

Detailed ways

Features and exemplary embodiments of various aspects of the invention will be described in detail below. The following description covers numerous specific details in order to provide a thorough understanding of the present invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced without some of these specific details. The following description of the embodiments is only to provide a clearer understanding of the present invention by showing examples of the present invention. The present invention is by no means limited to any specific configuration and algorithm presented below, but covers any modification, replacement and improvement of related elements, components and algorithms without departing from the spirit of the present invention.

<1. Gesture recognition system>

Figure 1 shows a block diagram of a general gesture recognition system. As shown in Figure 1, the gesture recognition system includes three functional parts: hand detection part, hand tracking part and gesture recognition part. Specifically, human motions are captured by a common web camera and the captured image frames are input into the gesture recognition system. For each image frame, first judge whether a hand object was detected in the previous frame. If the hand object has not been detected in the previous frame, the hand detection part will detect the hand object in the current whole frame. If a hand object has been detected in the previous frame, the hand tracking part will track the hand object in the current frame and the next frame. At the same time, the hand recognition part will analyze the historical traces to determine whether a predetermined gesture is recognized. If a predetermined gesture is recognized, the corresponding gesture command will be output for subsequent use, for example, to trigger a certain application.

As mentioned above, if a hand object has been detected in the previous frame, the hand tracking part will track the hand object in the current frame and the next frame. It can be seen that the calculation amount involved in tracking the hand object and the tracking accuracy are very important for the gesture recognition system. However, as mentioned above, the existing hand tracking methods require a large amount of calculation and the tracking accuracy is not satisfactory, especially when the hand moves through some objects with a color similar to the skin color, it is likely Tracking errors will occur due to occlusion. For example, in the traditional template matching algorithm, the hand target obtained in the previous frame is always used as the template for the template matching algorithm in the current frame, so when the hand moves through some objects with a color similar to the skin color Tracking errors can occur when hand target templates are misused or improperly used due to occlusions in the case of .

For example, Figure 2 shows an example where a tracking error occurs when a hand moves past a face. As shown in FIG. 2, five consecutive image frames 331-335 are shown in the figure. In image frames 331-333, the normal template matching algorithm is performed, that is, the hand tracking in image frame 331 is based on the previous frame The hand target detected or tracked in the image frame is used as a template, the hand tracking in the image frame 322 is based on the hand target tracked in the image frame 311 as a template, and the hand tracking in the image frame 333 is based on the hand target tracked in the image frame 332 as a template.

However, as shown in the figure, since the hand and the face overlap in the image frame 333, the hand object obtained in the image frame 333 contains a part of the human face with the same skin color, that is, the obtained hand The target is inaccurate or wrong. Therefore, when performing hand tracking for the image frame 334, the inaccurate or wrong hand target obtained in the image frame 333 is used as a template, and in this image frame 334, the false tracking of the hand continuing to stay on the face will be obtained through template matching. result. If this continues, wrong tracking results will also be obtained in the next image frame. It can be seen that if a tracking error occurs in one image frame, this error will cause a corresponding tracking error in the next image frame, so the tracking result is inaccurate, thereby affecting the correct recognition of gestures.

In view of the above-mentioned problems, the present invention provides a hand tracking method utilizing an adaptive template matching algorithm, which adaptively selects an appropriate template to perform a template matching algorithm by judging whether occlusion occurs in a frame in front of the current frame, High tracking accuracy can thereby be obtained even in the case where the hand moves past some object having a color similar to the skin color.

<2. Hand tracking method>

The hand tracking method according to the present invention will be described below with reference to FIG. 3 .

Fig. 3 shows a flowchart of the hand tracking method according to the present invention. As shown in FIG. 3, for each image frame, steps S301-S303 are performed. Specifically, in step S301, it is determined whether occlusion occurs in a frame preceding the current frame. In step S302, according to the occlusion judgment result, a hand object is adaptively selected from the hand objects tracked in the previous frame as a template. If occlusion occurs in the frame before the current frame, the hand object obtained in the frame immediately before the occlusion is used as the template; if no occlusion occurs in the frame before the current frame, the The obtained hand object is used as the template. Next, in step S303, a template matching method is performed for the current frame based on the template to obtain the hand object of the current frame. By using the hand tracking method of the present invention, an appropriate template can be selected according to whether occlusion occurs to execute a template matching algorithm, thereby avoiding false tracking caused by occlusion.

Next, the hand tracking method according to the present invention will be described in more detail with reference to FIG. 4 .

Fig. 4 shows a more detailed flowchart of the hand tracking method according to the present invention. As shown in FIG. 4 , first, in step S401 , the current skin color image and the current skin color difference image representing the motion difference between the current frame skin color image and the previous frame skin color image are calculated.

Next, in step S402, the search range is adaptively defined. The definition of the search range includes the following two steps: (1) In the skin color image of the current frame, the initial search is defined by expanding the surrounding area of the hand target tracked in the previous frame by a predetermined number of pixels (for example, 15 pixels). range; and (2) narrowing the initial search range to the smallest rectangle containing the hand target tracked in the previous frame and the current skin color difference image. In this way, the search scope used in performing the template matching method is obtained.

Next, in step S403, it is judged whether occlusion occurs in the frame preceding the current frame. Afterwards, according to the occlusion judgment result, a hand target is selected from the hand targets tracked in the frame before the current frame as a template for the template matching method in the current frame.

Specifically, as described in FIG. 3, if it is determined in step S403 that no occlusion has occurred in the frame preceding the current frame, ie, the determination result in step S403 is "No", the process proceeds to step S405. In step S405, like the traditional template matching algorithm, the hand object tracked in the previous frame of the current frame is used as a template for the template matching method in the current frame. At this time, the correction factor k of the motion difference is a fixed predetermined value, such as k=0.3, where the correction factor k is used in the template matching method performed later.

On the other hand, if it is determined in step S403 that occlusion has occurred in the frame preceding the current frame, that is, the result of determination in step S403 is "Yes", the procedure proceeds to step S404. In step S404, the hand object tracked in the frame immediately before the occlusion occurs is used as a template for the template matching method in the current frame. At this time, the correction factor k of the motion difference is calculated as the sum of the predetermined value and the percentage of PreOccDiff, for example, k=0.4+PreOccDiff/100, where PreOccDiff represents the skin color difference image of the frame immediately before the occlusion occurs and its previous frame The average value of each pixel in the hand object obtained in . For the occlusion judgment in the current frame, it will be introduced in a later step, and the occlusion judgment result for each frame will be stored for direct use in the next frame to judge whether occlusion occurred in the previous frame .

After the search range is defined in step S402 and a template is selected in step S404 or 405, in step S406, a template matching method is executed. In the present invention, the template matching method is a template matching method with motion difference correction. The key of the present invention is to select a template based on the occlusion judgment result, and to execute a template matching algorithm based on the thus selected template. Therefore, based on other considerations, a template matching algorithm without motion difference correction can also be used.

Specifically, in this template matching method, a plurality of first candidate hand ranges are defined in the defined search range, each first candidate hand range has the same size as the selected template, and the current skin color difference A second candidate hand range is defined in the image, and the second candidate hand range also has the same size as the selected template. In an exemplary embodiment, the hand target tracked in each frame and the candidate hand ranges defined in the search range and the skin color difference image are all rectangles in order to simplify the calculation, and the position of the rectangle is represented by the upper left of the rectangle corner vertices to update the position of the hand target with a uniform rule.

In the template matching method, the matching judgment process is cyclically executed for the multiple first candidate hand ranges until all the first candidate hand ranges have been matched and judged, so as to determine the candidate hand that best matches the selected template. range as the hand target tracked in the current frame, the matching judgment process includes the following steps:

Calculate the mean value of the absolute difference of the value of each pixel of the range of a first candidate hand and the selected template as the first error;

If the first error is greater than a first predetermined threshold, the current first candidate hand range is determined not to match the selected template and thus excluded;

If the first error is smaller than the first predetermined threshold, a second error is calculated, and the second error is obtained by subtracting the first error from the average value of each pixel value of the second candidate hand range multiplied by the correction factor. difference;

If the second error is smaller than the second predetermined threshold, the current first candidate hand range is determined to match the selected template, and the value of the second error and the average value of the values of the pixels in the current first candidate hand range are respectively As the second predetermined threshold in the matching determination process for the next first candidate hand range and the average value of the values of the pixels of the second candidate hand range.

Next, in step S407, it is judged based on the result of the template matching algorithm executed in step S406 whether any candidate hand range matches the selected template. Afterwards, based on the determination result and the information of the previous frame, it is determined whether occlusion occurs in the current frame and relevant parameters are updated for processing in the next frame. Specifically, if it is determined in step S407 that no candidate hand range matches the selected template, the process proceeds to step S409, and it is directly determined in step S409 that no occlusion occurs in the current frame. If it is determined in step S407 that a candidate hand range matches the selected template, the process proceeds to step S408, and occlusion condition judgment is performed in step S408 to determine whether occlusion occurs in the current frame.

The occlusion condition judgment performed in step S408 includes judging whether the following conditions are satisfied: (((PreOccDiff<=A)&&(DeltaDiff>B))||((PreOccDiff>A)&&(DeltaDiff>C)))&&(OccNum <D) Among them, PreOccDiff represents the average value of each pixel in the hand object tracked in the skin color difference image of the frame immediately before the occlusion and the previous frame, DeltaDiff is the difference between PreOccDiff and AvgTarDiff, and AvgTarDiff is the current skin color The average value of the values of each pixel in the range of the candidate hand that best matches the selected template in the difference image, OccNum represents the number of consecutive occlusions in front of the current frame, where C is greater than B, and where, if the above conditions are met, then in In step S410 it is determined that occlusion has occurred in the current frame, and if the above condition is not satisfied, in step S410 it is determined that occlusion has not occurred in the current frame.

In the above formula, the thresholds A-D are set based on experience. In situations such as a hand moving across a face, typically the occlusion will occur within about two frames of an image taken with a normal camera, so for example the threshold D could be set to 2, depending of course on the frame rate used by the camera and the speed of the hand movement and other factors, the threshold D can also be set to other suitable values. In addition, in fact, PreOccDiff reflects the intensity of hand movement, and the larger the PreOccDiff, the faster the hand movement or the stronger the hand movement, and vice versa. Therefore, among the values of B and C, C can take a larger value than B. For example, in the case where the threshold A both takes 15, B and C can be taken as B= and C=7 respectively. The specific example given here, ie A=15, B=4, C=5 and D=2, is used in the example given later in FIG. 4 .

Afterwards, in step S411 and step S412, the relevant parameters and the tracking track are updated according to the results of step S409 and step S410. Specifically, in the case that the hand target is successfully tracked in the current frame, if it is determined that no occlusion occurs in the current frame, the hand target tracked in the current frame is used as a template for the template matching method in the next frame. If masking occurs in the current frame, the template used for the template matching method in the current frame is still used for the template for the template matching method in the next frame. In addition, in the case that the hand object is not successfully tracked in the current frame, the hand object tracked in the frame before the occlusion is still used as the template for the template matching method in the next frame.

In the case of successfully tracking the hand target in the current frame, if it is determined that no occlusion occurs in the current frame, the tracking trajectory is updated as: P _N (x, y) = P (x, y), where P _N (x ,y) represents the coordinate position of the upper left corner of the rectangular hand target tracked in the current frame, and P(x,y) represents the position of the upper left corner of the candidate hand range that best matches the selected template. That is to say, the position of the upper left corner of the hand target tracked in the current frame is the position of the upper left corner of the candidate hand range determined to best match the selected template. If it is determined that occlusion occurs in the current frame, the tracking track is updated as:

where P _N-1 (x, y) represents the position of the upper left corner of the hand object tracked in the previous frame of the current frame, and P _N-2 (x, y) the hand tracked in the second frame before the current frame The position of the upper left corner of the target. In this way, the tracking process in the current frame ends, and then the next frame proceeds. The above method is executed for each frame in this way until all the frames are processed, so as to obtain the position of the tracked hand object in each frame so as to recognize the movement of the hand.

FIG. 5 shows an illustration of the tracking result of the hand tracking method when the hand moves in front of the face according to the present invention. In Fig. 5, the upper frame array is the skin color difference image between each frame and its previous frame, and for each frame also shows the AvgTargetDiff value to illustrate the effect of occlusion, this value can be used in step S408 Determine whether the masking condition is satisfied. Also, the frame array below shows the reference order of updating the templates used for the tracking process. As shown in FIG. 5 , occlusion is determined to occur in frame 80 , so the template for the template matching method in subsequent frame 81 is based on the tracked hand object in frame 79 before occlusion rather than based on tracking in frame 80 erroneous hand targets detected, thereby improving tracking accuracy even when the hand moves past an object (in this example, a human face) that has a color similar to the skin tone.

The hand tracking method using an adaptive template matching algorithm according to the present invention is not affected by brightness changes, and is not affected by objects with a color similar to skin color, and has low computational complexity to meet the requirements of implementing low-power devices Require.

The hand tracking method according to the present invention can be implemented in software, hardware, firmware and other ways. For example, when the hand tracking method is implemented in the form of software, the program for realizing the hand tracking method can be embedded in a storage medium, loaded and executed by a computer to implement the present invention. When the hand tracking method according to the present invention is implemented in hardware, the hand tracking method can be completed by corresponding components. For example, such a hand tracking device can be implemented, including: judging means for judging whether occlusion occurs in the frame before the current frame; selecting a hand target as a template; and obtaining means for performing a template matching method on the current frame based on the template to obtain the hand target of the current frame.

In addition, the hand tracking method according to the present invention can be used alone, and can also be used as a part of the gesture recognition process. Similarly, the hand tracking method device according to the present invention can be used alone, and can also be combined into a gesture recognition system as a component .

In this specification, the sequence in the given flow chart is described, but the hand tracking method of the present invention can also be implemented in other sequences, for example, some steps can be executed in parallel or a certain step can be performed in front of another step Execution, as long as the main idea of the present invention is satisfied, that is, for each frame, a template is adaptively selected according to the judgment of whether occlusion occurs in the previous frame to perform the template matching algorithm.

The hand tracking method and the hand tracking device according to the present invention are not limited to the above-mentioned uses, but can be applied to any occasion related to hand tracking.

Although the present invention has been described above with specific embodiments, those skilled in the art should understand that various modifications, combinations, sub-combinations and replacements may occur depending on design requirements and other factors, as long as they fall within the scope of the appended claims and their equivalents within the scope of the object.

Claims (9)

1. A hand tracking method, comprising: Determine whether occlusion occurs in the frame before the current frame; Adaptively select a hand target as a template from the hand targets tracked in previous frames according to the occlusion judgment result; and Carrying out a template matching method for the current frame based on the template to obtain the hand target of the current frame, Among them, adaptively selecting the hand target as a template includes: If an occlusion occurs in a frame preceding the current frame, the hand object obtained in the frame immediately before the occlusion occurs is used as the template; If no occlusion occurs in the previous frame of the current frame, the hand object obtained in the previous frame of the current frame is used as the template. 2. The hand tracking method according to claim 1, wherein the template matching method comprises: Define the search range and multiple candidate hand ranges in the skin color image of the current frame; Based on the search range and the template, find a candidate hand range in the search range that best matches the template as the hand target tracked in the current frame. 3. The hand tracking method according to claim 2, wherein defining the search range comprises: In the skin color image of the current frame, an initial search range is defined by enlarging the periphery of the hand target tracked in the previous frame by a predetermined number of pixels; and The initial search range is narrowed to the minimum rectangle containing the hand target tracked in the previous frame and the current skin color difference image representing the motion difference between the skin color image of the current frame and the skin color image of the previous frame as The search scope. 4. The hand tracking method according to claim 2 or 3, wherein the template matching method further comprises: A plurality of first candidate hand ranges are defined in the search range, the first candidate hand ranges have the same size as the template, and the distance between the skin color image representing the current frame and the skin color image of the previous frame is defined. A second candidate hand range is defined in the current skin color difference image of the motion difference, and the second candidate hand range has the same size as the template, The matching judgment process is cyclically executed for the plurality of first candidate hand ranges until the plurality of first candidate hand ranges have passed the matching judgment process, so as to determine the candidate hand range that best matches the template as the tracking frame in the current frame. Arrived hand target, The matching judgment process includes: Calculating the average value of the absolute difference between the first candidate hand range and the value of each pixel of the template as the first error; If the first error is greater than a first predetermined threshold, the current first candidate hand range is determined not to match the template and thus excluded; If the first error is less than the first predetermined threshold, a second error is calculated by subtracting the first error from the average value of the values of the pixels in the second candidate hand range The difference obtained by multiplying the value obtained by the correction factor; If the second error is smaller than the second predetermined threshold, the current first candidate hand range is determined to match the template, and the value of the second error and the average of the values of each pixel in the current first candidate hand range The values are taken as the second predetermined threshold and the average value of the values of the pixels of the second candidate hand range in the matching judgment process for the next first candidate hand range, respectively. 5. The hand tracking method according to claim 4, wherein, in a case where occlusion does not occur in a frame preceding the current frame, the correction factor is a first predetermined value, and if occlusion occurs in a frame preceding the current frame In this case, the correction factor is the sum of the second predetermined value and the percentage of the average value of the value of each pixel in the hand object obtained in the skin color difference image of the frame immediately before the occlusion occurs and the previous frame. 6. The hand tracking method according to claim 1, further comprising: judging whether occlusion occurs in the current frame, said judging comprising: If the hand target is not tracked in the current frame, it is directly determined that no occlusion occurs in the current frame; and If a hand object is tracked in the current frame, an occlusion condition judgment is performed to determine whether occlusion occurs in the current frame. 7. The hand tracking method according to claim 6, wherein said occlusion condition judgment includes judging whether the following conditions are met: (((PreOccDiff<=A)&&(DeltaDiff>B))||((PreOccDiff>A)&&(DeltaDiff>C)))&&(OccNum<D) among them, PreOccDiff means that the frame immediately before the occlusion occurs and the previous frame The average value of the value of each pixel in the hand object obtained in the skin color difference image of one frame, DeltaDiff is the difference between PreOccDiff and AvgTarDiff, and AvgTarDiff is each pixel in the current skin color difference image that best matches the template candidate hand The average value of the value, OccNum indicates the number of consecutive occlusions in front of the current frame, where C is greater than B, And wherein, if the above condition is satisfied, it is determined that occlusion has occurred in the current frame, and if the above condition is not satisfied, it is determined that occlusion has not occurred in the current frame. 8. A hand tracking device comprising: judging means, judging whether occlusion occurs in the frame before the current frame; The selection device adaptively selects a hand target as a template from the hand targets tracked in the previous frame according to the occlusion judgment result; and The obtaining device performs a template matching method for the current frame based on the template to obtain the hand target of the current frame, Wherein, adaptively selecting the hand target as a template includes: if occlusion occurs in the frame before the current frame, the hand target obtained in the frame immediately before the occlusion occurs is used as the template; If no occlusion occurs, the hand object obtained in the previous frame of the current frame is used as the template. 9. A gesture recognition system comprising the hand tracking device of claim 8.