TWI772208B - Method for counting number of people based on mmwave radar - Google Patents
Method for counting number of people based on mmwave radar Download PDFInfo
- Publication number
- TWI772208B TWI772208B TW110139192A TW110139192A TWI772208B TW I772208 B TWI772208 B TW I772208B TW 110139192 A TW110139192 A TW 110139192A TW 110139192 A TW110139192 A TW 110139192A TW I772208 B TWI772208 B TW I772208B
- Authority
- TW
- Taiwan
- Prior art keywords
- wave radar
- millimeter
- feature points
- point cloud
- people based
- Prior art date
Links
Images
Landscapes
- Radar Systems Or Details Thereof (AREA)
Abstract
Description
本發明涉及一種人數偵測方法,特別是涉及一種基於毫米波雷達的人數偵測方法。The present invention relates to a number of people detection method, in particular to a number of people detection method based on millimeter wave radar.
目前,對於商場或展覽館等室內空間來說,為了加強防疫,避免人數過多所發生的危險和不必要的傷害,人流的計數及掌握顯得相對重要許多。現有技術中,利用人數偵測以及室內定位等技術打造個人化需求的智能空間已能夠實現,例如利用攝影機或WIFI ®等。然而,WIFI ®通常需要多台接收與發射裝置,而攝影機則容易發生侵犯隱私的疑慮。此外,現有技術中有利用機器學習等技術來進行室內空間的人數偵測與定位。但利用此類技術需要蒐集大量資料以訓練模型,相當耗費人力與時間。 At present, for indoor spaces such as shopping malls or exhibition halls, in order to strengthen epidemic prevention and avoid danger and unnecessary injuries caused by excessive numbers of people, it is relatively important to count and control the flow of people. In the prior art, it has been possible to use technologies such as people detection and indoor positioning to create a smart space with personalized needs, such as using cameras or WIFI ® . However, WIFI ® usually requires multiple receivers and transmitters, and cameras are prone to privacy concerns. In addition, in the prior art, technologies such as machine learning are used to detect and locate people in indoor spaces. However, using such technology requires collecting a large amount of data to train the model, which is labor-intensive and time-consuming.
故,如何在人數計算的方法上進行適當改良,來克服上述的缺陷,已成為該領域所欲解決的重要課題之一。Therefore, how to properly improve the method of counting the number of people to overcome the above-mentioned defects has become one of the important issues to be solved in this field.
本發明所要解決的技術問題在於,針對現有技術的不足提供一種基於毫米波雷達的人數偵測方法。The technical problem to be solved by the present invention is to provide a number of people detection method based on millimeter wave radar in view of the deficiencies of the prior art.
為了解決上述的技術問題,本發明所採用的其中一技術方案是提供一種基於毫米波雷達的人數偵測方法,包括以下步驟: 利用毫米波雷達在一預設時間範圍內對預定空間內的多個移動中的人體發射出n個訊號,n個訊號通過多個移動中的人體的反射而由毫米波雷達接收,使毫米波雷達對應輸出n個特徵點,且n個特徵點集合成一幀點雲;累計三幀點雲中所包含的特徵點的數量,並且依據一搜索半徑以及一最小包含點數值,將每一幀點雲中的特徵點分類為有效特徵點與無效特徵點;以及依據一調整權重後的歐式距離公式將N幀點雲在預定空間內的分佈範圍集中,以將所有的有效特徵點區分出多個群體,且群體的數量代表在預定空間內的人數。In order to solve the above-mentioned technical problems, one of the technical solutions adopted by the present invention is to provide a method for detecting people based on a millimeter-wave radar, which includes the following steps: A moving human body emits n signals, and the n signals are received by the millimeter-wave radar through the reflection of multiple moving human bodies, so that the millimeter-wave radar outputs n feature points correspondingly, and the n feature points are collected into a frame of points. cloud; accumulating the number of feature points contained in three frames of point clouds, and classifying the feature points in each frame of point cloud into valid feature points and invalid feature points according to a search radius and a minimum included point value; and A weighted Euclidean distance formula concentrates the distribution range of N frames of point clouds in a predetermined space to distinguish all valid feature points into multiple groups, and the number of groups represents the number of people in the predetermined space.
本發明的其中一有益效果在於,本發明所提供的基於毫米波雷達的人數偵測方法,其能通過“依據一搜索半徑以及一最小包含點數值,將每一幀點雲中的特徵點分類為有效特徵點與無效特徵點”以及“依據一調整權重後的歐式距離公式將N幀點雲在預定空間內的分佈範圍集中,以將所有的有效特徵點區分出多個群體,且群體的數量代表在預定空間內的人數”的技術方案,以偵測出室內空間中的人數並且提高人數偵測的正確率。One of the beneficial effects of the present invention is that the method for detecting people based on millimeter-wave radar provided by the present invention can classify the feature points in each frame of point cloud by "according to a search radius and a minimum included point value. For the valid feature points and invalid feature points" and "According to a weighted Euclidean distance formula, the distribution range of N frames of point clouds in the predetermined space is concentrated, so as to distinguish all valid feature points into multiple groups, and the group's distribution range. The number represents the number of people in the predetermined space" technical solution to detect the number of people in the indoor space and improve the accuracy of the number of people detection.
為使能更進一步瞭解本發明的特徵及技術內容,請參閱以下有關本發明的詳細說明與圖式,然而所提供的圖式僅用於提供參考與說明,並非用來對本發明加以限制。For a further understanding of the features and technical content of the present invention, please refer to the following detailed descriptions and drawings of the present invention. However, the drawings provided are only for reference and description, and are not intended to limit the present invention.
以下是通過特定的具體實施例來說明本發明所公開有關“基於毫米波雷達的人數偵測方法”的實施方式,本領域技術人員可由本說明書所公開的內容瞭解本發明的優點與效果。本發明可通過其他不同的具體實施例加以施行或應用,本說明書中的各項細節也可基於不同觀點與應用,在不背離本發明的構思下進行各種修改與變更。另外,本發明的附圖僅為簡單示意說明,並非依實際尺寸的描繪,事先聲明。以下的實施方式將進一步詳細說明本發明的相關技術內容,但所公開的內容並非用以限制本發明的保護範圍。另外,應當可以理解的是,雖然本文中可能會使用到“第一”、“第二”、“第三”等術語來描述各種元件,但這些元件不應受這些術語的限制。這些術語主要是用以區分一元件與另一元件。另外,本文中所使用的術語“或”,應視實際情況可能包括相關聯的列出項目中的任一個或者多個的組合。另外,本文中所使用的術語“或”,應視實際情況可能包括相關聯的列出項目中的任一個或者多個的組合。The following is a specific embodiment to illustrate the implementation of the "people detection method based on millimeter wave radar" disclosed in the present invention, and those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the concept of the present invention. In addition, the drawings of the present invention are merely schematic illustrations, and are not drawn according to the actual size, and are stated in advance. The following embodiments will further describe the related technical contents of the present invention in detail, but the disclosed contents are not intended to limit the protection scope of the present invention. Additionally, it should be understood that, although the terms "first," "second," "third," etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are primarily used to distinguish one element from another. In addition, the term "or", as used herein, should include any one or a combination of more of the associated listed items, as the case may be. In addition, the term "or", as used herein, should include any one or a combination of more of the associated listed items, as the case may be.
[實施例][Example]
參閱圖1所示,圖1為本發明的基於毫米波雷達的人數偵測方法的運用場景示意圖,讓受測者在一室內空間中進行一段時間(例如5分鐘左右)的自由移動。需先說明的是,本發明並不對受測者的人數作限制,並且受測者的移動軌跡也不限,例如可以是來回走動、繞圈走動或是8字型走動等等,另外,移動的速度也不限。接著,在該室內空間中的天花板設置一毫米波雷達1,用以該室內空間的人數及移動軌跡。此外,在該室內空間中還設置至少一紅外線感測器2,舉例來說,在本實施例中是以兩位受測者為例,因此紅外線感測器2也對應設置兩台(分別位於受測者前方及側邊)來分別偵測兩位受測者的實際軌跡。毫米波雷達1距離地面約2.9公尺,而每一紅外線感測器2距離地面約0.85公尺且距離對應的受測者約3公尺。本發明所提供的人數偵測方法主要通過毫米波雷達1來實現,並且通過紅線感測器2偵測兩位受測者的實際軌跡來驗證該方法的準確度。Referring to FIG. 1 , FIG. 1 is a schematic diagram of an application scenario of the millimeter-wave radar-based people detection method of the present invention, allowing the subject to move freely in an indoor space for a period of time (eg, about 5 minutes). It should be noted that the present invention does not limit the number of subjects, and the movement trajectory of the subjects is not limited. speed is not limited. Next, a millimeter-wave radar 1 is installed on the ceiling of the indoor space for the number of people and movement trajectories in the indoor space. In addition, at least one
承上述,毫米波雷達1是基於頻率原理的調頻連續波雷達,一般具有一發射端與一接收端。毫米波雷達1能夠同時由該發射端發射出多個調頻連續波,並且經過受測者的身體不同部位反射後再由該接收端接收。因此,由發射/接收的訊號可以對應生成多個不同相位的中頻訊號(Imtermediate Frequency, IF),使得毫米波雷達1據以計算且產生出受測者的身體不同部位的特徵點。每一個訊號能夠生成一個特徵點,而毫米波雷達1能夠在每一幀輸出多個特徵點,並且該多個特徵點可集合成一個點雲。在本發明的實施例中,一幀是代表0.1秒,因此一幀點雲指的是在0.1秒的時間內所形成的一個點雲。值得一提的是,毫米波雷達1只對運動的物體進行偵測,因此只有當受測者動作時才會收集到點雲。另外需說明的是,多個特徵點的生成次序並非是按照和毫米波雷達1的距離遠近無關,且毫米波雷達1在每一幀輸出的點雲所包含的特徵點個數也是不固定。Based on the above, the millimeter-wave radar 1 is a frequency-modulated continuous wave radar based on the frequency principle, and generally has a transmitter and a receiver. The millimeter-wave radar 1 can simultaneously transmit a plurality of frequency-modulated continuous waves from the transmitting end, which are reflected by different parts of the subject's body before being received by the receiving end. Therefore, a plurality of intermediate frequency (IF) signals with different phases can be generated correspondingly from the transmitted/received signals, so that the millimeter wave radar 1 can calculate and generate the feature points of different parts of the subject's body accordingly. Each signal can generate one feature point, and the millimeter wave radar 1 can output multiple feature points in each frame, and the multiple feature points can be aggregated into a point cloud. In the embodiment of the present invention, one frame represents 0.1 second, so one frame of point cloud refers to a point cloud formed within 0.1 second. It is worth mentioning that the millimeter-wave radar 1 only detects moving objects, so the point cloud will only be collected when the subject moves. In addition, it should be noted that the order in which the multiple feature points are generated is not independent of the distance of the millimeter-wave radar 1 , and the number of feature points included in the point cloud output by the millimeter-wave radar 1 in each frame is also not fixed.
承上述,每一個特徵點包含多個不同的資訊,具體來說,特徵點包含一預設距離R、一水平角θ、一仰角φ、一速度V以及一信噪比(SNR)等五個特徵。參閱圖2所示,圖2為本發明的基於毫米波雷達的人數偵測方法的運用場景的座標系轉換示意圖,由於每個特徵點皆具有的R、θ、 ,因此將圖1中的室內空間轉換為直角座標系,其中,其中毫米波雷達1為原點(0,0,0),而每一特徵點的坐標則為(x,y,z),而 , , 。預設距離R為每一特徵點與毫米波雷達1之間的直線距離,水平角θ為預設距離R在XY平面上的投影距離與Y軸之間的夾角,仰角 為預設距離R在XY平面上的投影距離與預設距離R之間的夾角,速度V為毫米波雷達1利用都卜勒效應的原理進行測量而得到的受測者在雷達波束方向上的徑向運動速度。藉此,利用毫米波雷達1偵測室內空間的受測者一段時間之後,將獲得多個點雲在直角坐標系上的分布,如圖3所示,圖3為點雲經過可視化調整後在直角坐標系的示意圖。 Based on the above, each feature point contains a plurality of different pieces of information. Specifically, the feature point includes five preset distances R, a horizontal angle θ, an elevation angle φ, a velocity V, and a signal-to-noise ratio (SNR). feature. Referring to FIG. 2, FIG. 2 is a schematic diagram of coordinate system conversion of the application scene of the millimeter-wave radar-based people detection method of the present invention, since each feature point has R, θ, , so the indoor space in Figure 1 is converted into a rectangular coordinate system, where the millimeter-wave radar 1 is the origin (0, 0, 0), and the coordinates of each feature point are (x, y, z), and , , . The preset distance R is the straight-line distance between each feature point and the millimeter-wave radar 1, the horizontal angle θ is the angle between the projection distance of the preset distance R on the XY plane and the Y axis, and the elevation angle is the angle between the projection distance of the preset distance R on the XY plane and the preset distance R, and the velocity V is the measured object’s velocity in the direction of the radar beam obtained by the millimeter-wave radar 1 using the Doppler effect principle. Radial movement speed. In this way, after using the millimeter-wave radar 1 to detect the subject in the indoor space for a period of time, the distribution of multiple point clouds in the rectangular coordinate system will be obtained, as shown in Figure 3. Figure 3 shows the point cloud after visual adjustment. Schematic diagram of the Cartesian coordinate system.
接著,參閱圖4所示,圖4為本發明的基於毫米波雷達的人數偵測方法所採用的Overlay-DBSCAN演算法的原理示意圖,其包括以下步驟:Next, referring to Fig. 4, Fig. 4 is a schematic diagram of the Overlay-DBSCAN algorithm adopted by the millimeter-wave radar-based people detection method of the present invention, which comprises the following steps:
步驟S100: 利用毫米波雷達1在一預設時間範圍內對預定空間內的多個移動中的人體發射出n個訊號,n個訊號通過多個移動中的人體的反射而由毫米波雷達1接收,使毫米波雷達1對應產生出n個特徵點,且n個特徵點可集合成一個點雲,在本實施例中,預設時間範圍為一幀,也就是0.1秒;Step S100: The millimeter-wave radar 1 is used to transmit n signals to a plurality of moving human bodies in a predetermined space within a preset time range, and the n signals are reflected by the millimeter-wave radar 1 through the reflection of the plurality of moving human bodies. receiving, so that the millimeter-wave radar 1 generates n feature points correspondingly, and the n feature points can be assembled into a point cloud. In this embodiment, the preset time range is one frame, that is, 0.1 seconds;
步驟S102: 累計三幀點雲中所包含的特徵點的數量,並且依據一搜索半徑以及一最小包含點數值,將每一幀點雲中的所有特徵點分類為有效特徵點與無效特徵點。Step S102: Accumulate the number of feature points included in the three frames of point clouds, and classify all feature points in each frame of point cloud into valid feature points and invalid feature points according to a search radius and a minimum included point value.
須說明的是,在本實施例中,搜索半徑為介於0.4公尺與0.52公尺之間。It should be noted that, in this embodiment, the search radius is between 0.4 meters and 0.52 meters.
步驟S104: 依據一調整權重後的歐式距離公式將N幀點雲在該預定空間內的分佈範圍集中,並且將所有的有效特徵點區分出多個群體,該群體的數量即代表在預定空間內的人數。Step S104: According to the Euclidean distance formula after adjusting the weight, the distribution range of N frames of point clouds in the predetermined space is concentrated, and all the effective feature points are distinguished into multiple groups, and the number of the groups is represented in the predetermined space. number of people.
本發明所採用的Overlay-DBSCAN演算法是種針對DBSCAN演算法進行改良的聚類演算法,主要用以發現數據間的分布規律和隱含關係,也就是把大量的N個數據樣本聚集成K個群體,使同一類的樣本相似性最大而不同類的樣本相似性最小。而在本發明的實施例中,即是將利用毫米波雷達1所收集到的N幀點雲依據室內空間中的人數而聚集成對應的群體數量(也就是說,聚集成K個群體代表著室內空間中有K個人)。進一步來說,由於毫米波雷達1所輸出的每一幀點雲代表一堆隨機數量且含有空間座標資訊的特徵點,而這些特徵點不僅包含因受測者動作而產生的有效特徵點,也有極少數是因受到外界環境的影響,例如:高頻的電磁訊號,變電箱,金屬外殼的電腦等等而產生的無效特徵點(稱之為雜訊點)。因此,為了確定哪些是有效特徵點而那些是無效特徵點,本發明採用DBSCAN演算法,其原理為在預設的一搜索半徑r(以公尺為單位)的點雲數量不小於一最小包含點數M,也就是在某個時段內,搜索半徑r內有最小包含點數M以上的特徵點數量的點雲集合,此外,搜索半徑r是在偵測過程中能夠動態調整,並非是固定不變。利用此方法,多個特徵點可分為核心點(Core points)、邊界點(Border points)以及雜訊點(Noise points)。舉例來說,如圖3所示,首先任意選取一個特徵點,以該特徵點為圓心並且以預設的一搜索半徑r畫圓,若在該搜索半徑r的範圍內的其他特徵點數量超過最小包含點數M,則稱該特徵點為核心點(如圖4中的點A);接著,繼續任意選取另一特徵點,並且以該特徵點為圓心且以另一搜索半徑r畫圓,若在該另一搜索半徑r內的其他特徵點數量小於最小包含點數M,並且這一做為圓心的特徵點落在核心點範圍內,那麼該特徵點則稱為邊界點(如圖3中的點B及點C)。至於不屬於以上兩者的就是雜訊點(如圖3中的點N)。由上述可知,聚類演算法的準確性與搜索半徑r及最小包含點數M兩個參數有關。若預設的搜索半徑r過大的會導致核心點數量減少,使一些原本應該包含在內的特徵點被丟棄;選擇過小的最小包含點數M則會導致大量的對象被標記為核心點,從而將雜訊點也當成是核心點,進而將雜訊點歸類成點雲集合,而給定最小包含點數卻選擇過小的搜索半徑r會導致大量的對象被當成雜訊點。The Overlay-DBSCAN algorithm adopted in the present invention is an improved clustering algorithm for the DBSCAN algorithm, which is mainly used to discover the distribution law and implicit relationship between data, that is, to gather a large number of N data samples into K group, so that the samples of the same class have the largest similarity and the samples of the same class have the smallest similarity. In the embodiment of the present invention, the N frames of point clouds collected by the millimeter-wave radar 1 are aggregated into the corresponding number of groups according to the number of people in the indoor space (that is, the aggregation into K groups represents the There are K persons in the indoor space). Further, since each frame of point cloud output by the millimeter-wave radar 1 represents a random number of feature points that contain spatial coordinate information, these feature points not only include valid feature points generated by the subject's actions, but also Very few are invalid feature points (called noise points) caused by the influence of the external environment, such as high-frequency electromagnetic signals, substations, computers with metal casings, etc. Therefore, in order to determine which are valid feature points and which are invalid feature points, the present invention adopts the DBSCAN algorithm, the principle of which is that the number of point clouds in a preset search radius r (in meters) is not less than a minimum containing The number of points M, that is, in a certain period of time, the search radius r has the minimum number of point clouds containing the number of feature points above the number of points M. In addition, the search radius r can be dynamically adjusted during the detection process, not fixed. constant. Using this method, multiple feature points can be divided into core points, border points and noise points. For example, as shown in Fig. 3, first select a feature point arbitrarily, draw a circle with the feature point as the center and a preset search radius r, if the number of other feature points within the range of the search radius r exceeds The minimum number of included points M is called the core point (point A in Figure 4); then, continue to arbitrarily select another feature point, and draw a circle with this feature point as the center and another search radius r , if the number of other feature points within the other search radius r is less than the minimum number of included points M, and this feature point as the center of the circle falls within the range of the core point, then the feature point is called a boundary point (as shown in the figure 3 points B and C). What does not belong to the above two is the noise point (point N in Figure 3). It can be seen from the above that the accuracy of the clustering algorithm is related to two parameters, the search radius r and the minimum number of included points M. If the preset search radius r is too large, the number of core points will be reduced, and some feature points that should be included will be discarded; if the minimum number of included points M is selected too small, a large number of objects will be marked as core points, so The noise points are also regarded as core points, and then the noise points are classified into a point cloud set, and a search radius r that is too small for a given minimum number of included points will cause a large number of objects to be regarded as noise points.
進一步來說,因為受測者的身高體重會導致身形有所差異,因此每個受測者被毫米波雷達1掃描後,回饋給毫米波雷達1的點雲數量一定不同,且因為人體的動作幅度大小也會造成點雲數量的改變。如果場景內人數增加,則點雲數量的差異會更加巨大,這也就造成了如果在聚類的時候,對每一幀點雲都使用同一最小包含點數的話,效果會非常有限。因此,需要動態調整每一幀點雲在搜索半徑r範圍內的最小包含點數M,來避免如果有某一幀點雲數量過少時,演算法會將所產生的點雲數量過少的另外一個人直接當成雜訊處理。本發明通過將每三幀的點雲所包含的特徵點的數量累加在一起,在搜索半徑r相同的情況下,將最小包含點數4、5、6、7四個值進行運算,並依據所獲得的結果從中選出一雜訊點數量最少者當作這三幀點雲最適合的最小包含點數M。之所以採用累加三幀點雲的方式進行計算,是由於一幀是0.1秒,人體在正常行走下在0.3秒內的變化不會太大。如此可以保證在0.3秒內人體的質心位置不會差異太多,且同時也可以避免上述提到的過多點雲被誤判為雜點的問題。Further, because the height and weight of the subjects will lead to differences in body shape, after each subject is scanned by the millimeter-wave radar 1, the number of point clouds fed back to the millimeter-wave radar 1 must be different, and because the human body The magnitude of the motion also changes the number of point clouds. If the number of people in the scene increases, the difference in the number of point clouds will be even greater, which also causes the effect to be very limited if the same minimum number of included points is used for each frame of point cloud during clustering. Therefore, it is necessary to dynamically adjust the minimum number of points M included in the search radius r of each frame of point cloud to avoid that if there is too little point cloud in a certain frame, the algorithm will generate another person whose number of point clouds is too small. Treat it directly as noise. In the present invention, by accumulating the number of feature points contained in point clouds of every three frames, under the condition that the search radius r is the same, the minimum number of contained points 4, 5, 6, and 7 is calculated, and based on From the obtained results, the one with the least number of noise points is selected as the most suitable minimum number M of points contained in the point clouds of these three frames. The reason why the calculation is performed by accumulating three frames of point clouds is that one frame is 0.1 seconds, and the human body does not change much within 0.3 seconds under normal walking. In this way, it can be ensured that the position of the center of mass of the human body will not be too different within 0.3 seconds, and at the same time, it can also avoid the above-mentioned problem of excessive point clouds being misjudged as noise points.
繼續參閱圖3所示,由圖3可知,點雲在X、Y、Z軸的分佈其實相當的不平均,在Y軸上分布較為鬆散。因此,本發明依據一調整權重後的歐式距離公式將N幀所述點雲在所述預定空間內的分佈範圍集中,以將所有的有效特徵點區分出多個群體,且所述群體的數量代表在所述預定空間內的人數。所述調整權重後的歐式距離公式即為本發明所採用的Overlay-DBSCAN演算法的公式,包括:Continuing to refer to Figure 3, it can be seen from Figure 3 that the distribution of point clouds on the X, Y, and Z axes is actually quite uneven, and the distribution on the Y axis is relatively loose. Therefore, the present invention concentrates the distribution range of N frames of the point cloud in the predetermined space according to a weight-adjusted Euclidean distance formula, so as to distinguish all valid feature points into multiple groups, and the number of the groups represents the number of people in the predetermined space. The Euclidean distance formula after the described adjustment weight is the formula of the Overlay-DBSCAN algorithm adopted by the present invention, including:
D(P i,P j)=(p xi-p xj) 2+k(p yi-p yj) 2+(p zi-p zj) 2....(1); D(P i ,P j )=(p xi -p xj ) 2 +k(p yi -p yj ) 2 +(p zi -p zj ) 2 ....(1);
其中,D(P i,P j)為兩個不同的所述特徵點 i(p xi,p yi,p zi)與P j(p xj,p yj,p zj)之間的距離,k為權重且範圍為0<k<1。較佳者,在本發明的實施例中,k為0.2(也就是在Y軸上分配較小的權重)。經過調整權重後的點雲可視化的座標圖如圖5所示,可看出經過調整權重後的點雲會壓縮而變得較為集中,有助於不管在直觀上區分得更精準,即更容易區分為兩個群體,更接近實際上的受測者數量(兩位)。 Wherein, D(P i , P j ) is the distance between two different feature points i (p xi , p yi , p zi ) and P j (p xj , p yj , p zj ), and k is weight and in the range 0<k<1. Preferably, in the embodiment of the present invention, k is 0.2 (ie, a smaller weight is assigned on the Y-axis). The coordinate diagram of the point cloud visualization after adjusting the weight is shown in Figure 5. It can be seen that the point cloud after adjusting the weight will be compressed and become more concentrated, which helps to distinguish more accurately regardless of the intuitiveness, that is, it is easier Divided into two groups, closer to the actual number of subjects (two).
然而,經過Overlay-DBSCAN演算法獲得圖5的點雲可視化座標圖在實際量測上,若兩位受測者移動的軌跡太近,容易產生歸類到群體甲的點雲被歸類到群體乙的情況,也就是發生點雲錯置的情況。在經過Overlay-DBSCAN演算法獲得圖5的點雲可視化座標圖之後,可進一步採用匈牙利演算法來降低點雲的錯置率。繼續參閱圖6所示,圖6為本發明的基於毫米波雷達的人數偵測方法的步驟示意圖,包括以下步驟:However, after the Overlay-DBSCAN algorithm is used to obtain the visual coordinate diagram of the point cloud in Figure 5, in actual measurement, if the moving trajectories of the two subjects are too close, it is easy to generate a point cloud classified into group A and classified into the group. The case of B, that is, the situation where the point cloud is misplaced. After the Overlay-DBSCAN algorithm is used to obtain the visualized coordinate map of the point cloud in Figure 5, the Hungarian algorithm can be further used to reduce the misplacement rate of the point cloud. Continue to refer to as shown in Figure 6, Figure 6 is a schematic diagram of the steps of the millimeter-wave radar-based people detection method of the present invention, comprising the following steps:
步驟S106: 依據匈牙利演算法計算出多個群體中的任一幀點雲的錯置率;Step S106: Calculate the dislocation rate of any frame point cloud in multiple groups according to the Hungarian algorithm;
步驟S108: 依據卡爾曼濾波器預測出多個移動中的人體的移動軌跡。Step S108: Predicting the movement trajectories of a plurality of moving human bodies according to the Kalman filter.
舉例來說,由於每一個點雲中包含了許多特徵點有關位置的三維座標數據,因此,若k時刻具有點雲數量例如為45的群體甲可具有一關於三維座標數據的矩陣如下:For example, since each point cloud contains many three-dimensional coordinate data related to the position of the feature points, if the number of point clouds at time k is 45, for example, group A can have a matrix related to the three-dimensional coordinate data as follows:
k時刻: k moment:
同樣地,k+1時刻具有點雲數量例如為80的一群體乙也可具有一關於三維座標數據的矩陣如下Similarly, a group B with a point cloud number of 80 at time k+1 can also have a matrix about the three-dimensional coordinate data as follows
k+1時刻: k+1 moment:
因此,經過匈牙利演算法的運算後得到兩兩之間的矩陣相關係數,也就是得到點雲數量為45的群體與點雲數量為80的群體之間的相關性,以避免產生歸類到群體甲的其中一點雲被歸類到群體乙的情況,也就是發生點雲錯置的情況。Therefore, after the operation of the Hungarian algorithm, the matrix correlation coefficient between each pair is obtained, that is, the correlation between the group with 45 point clouds and the group with 80 point clouds is obtained, so as to avoid classifying into groups. One of the point clouds of A is classified into the case of group B, that is, the situation where the point cloud is misplaced.
須說明的是,會採用相關係數的原因為在第k時刻到第k+1時刻的這段時間,每個人移動的距離不會差距太大,而相關係數又可以度量兩個群體之間的相關程度,故採用相關係數來計算第k時刻到第k+1時刻的相關性。接著,依據所述相關矩陣相關係數可逐一確認每一個群體中點雲是否被錯誤歸類(即錯置)。It should be noted that the reason for using the correlation coefficient is that during the period from the kth time to the k+1th time, the distance moved by each person will not be too different, and the correlation coefficient can measure the relationship between the two groups. Therefore, the correlation coefficient is used to calculate the correlation from the kth time to the k+1th time. Then, according to the correlation coefficient of the correlation matrix, it can be confirmed whether the point cloud in each group is wrongly classified (ie, misplaced).
進一步來說,經過Overlay-DBSCAN演算法的聚類以及匈牙利演算法的固定身份之後,可再利用卡爾曼濾波器進行人移動的軌跡追蹤,使得由毫米波雷達1預測獲得的軌跡能夠接近於紅外線感測器2測量到的實際觀測軌跡。卡爾曼濾波器是一種最佳化回歸數據處理演算法,利用上一刻的狀態估計值與當前狀態的觀測值來對下一狀態做預測值。Further, after the clustering of the Overlay-DBSCAN algorithm and the fixed identity of the Hungarian algorithm, the Kalman filter can be used to track the trajectory of human movement, so that the trajectory predicted by the millimeter-wave radar 1 can be close to the infrared ray. The actual observation track measured by the
最佳化回歸數據演算公式包括以下關係式:The optimal regression data calculation formula includes the following relations:
X k|k-1=AX k-1|k-1+Bu k-1; X k|k-1 =AX k-1|k-1 +Bu k-1 ;
P k|k-1=AP k-1|k-1A T+Q; P k|k-1 =AP k-1|k-1 A T +Q;
X k|k=X k|k-1+K k(Z k‒HX k|k-1); X k|k =X k|k-1 +K k (Z k ‒HX k|k-1 );
K k= P k|k-1H T(HP k|k-1H T+HX k|k-1) -1; K k = P k|k-1 H T (HP k|k-1 H T +HX k|k-1 ) -1 ;
P k|k=(I‒ K kH)P k|k-1。 P k|k =(I‒ K k H)P k|k-1 .
其中,X
k|k為第k時刻的系統狀態,X
k|k-1為上一狀態預測的結果,也就是毫米波雷達1藉由第k-1時刻預測第k時刻狀態的估計值。A跟B為系統參數,假設系統為多參數模型系統時,A 和 B 為矩陣,A
T為A的轉置矩陣。Z
k為第k時刻的測量值,也就是紅外線感測器2實際測量到的觀測值,H為一對應估計值的矩陣參數,K
k代表估計值與觀測值之間的權重。P 表示相對於系統狀態 x 的共變異數。K為增益值(Kalman Gain)。H 是測量系統的參數,是指過程中和測量所得到的雜訊,他們的共變異數分別是 Q,R。
Among them, X k|k is the system state at the k-th time, and X k|k-1 is the result of the previous state prediction, that is, the millimeter-wave radar 1 predicts the k-th state at the k-1 time. The estimated value of the state. A and B are system parameters. If the system is a multi-parameter model system, A and B are matrices, and A T is the transpose matrix of A. Z k is the measured value at the kth time, that is, the observed value actually measured by the
[實施例的有益效果][Advantageous effects of the embodiment]
本發明的其中一有益效果在於,本發明所提供的基於毫米波雷達的人數偵測方法,其能通過“依據一搜索半徑以及一最小包含點數值,將每一幀點雲中的特徵點分類為有效特徵點與無效特徵點”以及“依據一調整權重後的歐式距離公式將N幀點雲在預定空間內的分佈範圍集中,以將所有的有效特徵點區分出多個群體,且群體的數量代表在預定空間內的人數”的技術方案,以偵測出室內空間中的人數並且提高人數偵測的正確率。One of the beneficial effects of the present invention is that the method for detecting people based on millimeter-wave radar provided by the present invention can classify the feature points in each frame of point cloud by "according to a search radius and a minimum included point value. are valid feature points and invalid feature points" and "the distribution range of N frames of point clouds in a predetermined space is concentrated according to a weight-adjusted Euclidean distance formula, so as to distinguish all valid feature points into multiple groups, and the group's The number represents the number of people in the predetermined space" technical solution to detect the number of people in the indoor space and improve the accuracy of the number of people detection.
更進一步來說,本發明利用毫米波雷達1所產生出的三維點雲進行室內定位,首先對於現有技術中的的聚類進行了改進,在預處理上先給予不同軸值不同的權重後,再以累加的方式解決點雲某些幀數數量過少的問題,接著輔以統計的方式算出最適合那幀的最小包含點數值。藉由本發明的方法所得到的實驗結果,人數準確率從調整前的45.53%進步至90.18%,濾除雜訊的效果更是改善了許多。接著,本發明的在利用匈牙利演算法使得在聚類後的結果在時間與空間上有了更強大的關聯性,讓每個人移動的軌跡結果在前後是具有相關的,兩人身份錯置率僅約為8%,最後利用卡爾曼濾波器的濾波之後,使得由毫米波雷達1預測獲得的軌跡接近於紅外線感測器2測量到的實際觀測軌跡。Furthermore, the present invention uses the three-dimensional point cloud generated by the millimeter-wave radar 1 to perform indoor positioning. First, the clustering in the prior art is improved. In preprocessing, different weights are given to different axis values. Then solve the problem that the number of frames in the point cloud is too small by means of accumulation, and then use a statistical method to calculate the minimum value of the minimum contained points that is most suitable for that frame. According to the experimental results obtained by the method of the present invention, the accuracy rate of the number of people is improved from 45.53% before adjustment to 90.18%, and the effect of filtering out noise is much improved. Then, the Hungarian algorithm of the present invention makes the clustering results have a stronger correlation in time and space, so that the trajectory results of each person's movement are related before and after, and the two people's identity misplacement rate It is only about 8%. Finally, after filtering by the Kalman filter, the trajectory predicted by the millimeter wave radar 1 is close to the actual observed trajectory measured by the
以上所公開的內容僅為本發明的優選可行實施例,並非因此侷限本發明的申請專利範圍,所以凡是運用本發明說明書及圖式內容所做的等效技術變化,均包含於本發明的申請專利範圍內。The contents disclosed above are only preferred feasible embodiments of the present invention, and are not intended to limit the scope of the present invention. Therefore, any equivalent technical changes made by using the contents of the description and drawings of the present invention are included in the application of the present invention. within the scope of the patent.
1:毫米波雷達 2:紅外線感測器 r:搜索半徑 A核心點 B、C:邊界點 N:雜訊點 S100、S102、S104、S106、S108:步驟1: Millimeter wave radar 2: Infrared sensor r: search radius A core point B, C: boundary points N: noise point S100, S102, S104, S106, S108: Steps
圖1為本發明的基於毫米波雷達的人數偵測方法的運用場景示意圖。FIG. 1 is a schematic diagram of an application scenario of the method for detecting people based on a millimeter-wave radar according to the present invention.
圖2為本發明的基於毫米波雷達的人數偵測方法的運用場景的座標系轉換示意圖。FIG. 2 is a schematic diagram of coordinate system conversion of an application scene of the millimeter-wave radar-based people detection method of the present invention.
圖3為點雲可視化後在直角坐標系的示意圖。Figure 3 is a schematic diagram of the point cloud after visualization in the Cartesian coordinate system.
圖4為本發明的基於毫米波雷達的人數偵測方法所採用的Overlay-DBSCAN演算法的原理示意圖。FIG. 4 is a schematic diagram of the principle of the Overlay-DBSCAN algorithm adopted by the millimeter-wave radar-based people detection method of the present invention.
圖5為經過調整後的點雲可視化後在直角坐標系的示意圖。FIG. 5 is a schematic diagram of the adjusted point cloud after visualization in the Cartesian coordinate system.
圖6為本發明的基於毫米波雷達的人數偵測方法的步驟示意圖。FIG. 6 is a schematic diagram of steps of a method for detecting a person based on a millimeter wave radar according to the present invention.
S100、S102、S104、S106、S108:步驟 S100, S102, S104, S106, S108: Steps
Claims (8)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW110139192A TWI772208B (en) | 2021-10-22 | 2021-10-22 | Method for counting number of people based on mmwave radar |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW110139192A TWI772208B (en) | 2021-10-22 | 2021-10-22 | Method for counting number of people based on mmwave radar |
Publications (2)
Publication Number | Publication Date |
---|---|
TWI772208B true TWI772208B (en) | 2022-07-21 |
TW202318031A TW202318031A (en) | 2023-05-01 |
Family
ID=83439809
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
TW110139192A TWI772208B (en) | 2021-10-22 | 2021-10-22 | Method for counting number of people based on mmwave radar |
Country Status (1)
Country | Link |
---|---|
TW (1) | TWI772208B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWI800471B (en) * | 2022-11-09 | 2023-04-21 | 元智大學 | Method for counting number of people based on mmwave radar |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW201830943A (en) * | 2017-01-10 | 2018-08-16 | 美商高通公司 | Co-existence of millimeter wave communication and radar |
CN109255386A (en) * | 2018-09-18 | 2019-01-22 | 中国科学院沈阳自动化研究所 | The road pedestrian rapid detection method merged based on millimetre-wave radar and vision |
CN111239728A (en) * | 2020-02-26 | 2020-06-05 | 深圳雷研技术有限公司 | Passenger counting method and system based on millimeter wave radar |
CN111538013A (en) * | 2020-05-12 | 2020-08-14 | 上海眼控科技股份有限公司 | Radar echo extrapolation method and device, computer equipment and storage medium |
EP3739356A1 (en) * | 2019-05-12 | 2020-11-18 | Origin Wireless, Inc. | Method, apparatus, and system for wireless tracking, scanning and monitoring |
CN112305576A (en) * | 2020-10-31 | 2021-02-02 | 中环曼普科技(南京)有限公司 | Multi-sensor fusion SLAM algorithm and system thereof |
TW202131013A (en) * | 2019-10-04 | 2021-08-16 | 美商高通公司 | Near field user detection using a radar signal analysis associated with a user equipment |
-
2021
- 2021-10-22 TW TW110139192A patent/TWI772208B/en active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW201830943A (en) * | 2017-01-10 | 2018-08-16 | 美商高通公司 | Co-existence of millimeter wave communication and radar |
CN109255386A (en) * | 2018-09-18 | 2019-01-22 | 中国科学院沈阳自动化研究所 | The road pedestrian rapid detection method merged based on millimetre-wave radar and vision |
EP3739356A1 (en) * | 2019-05-12 | 2020-11-18 | Origin Wireless, Inc. | Method, apparatus, and system for wireless tracking, scanning and monitoring |
TW202131013A (en) * | 2019-10-04 | 2021-08-16 | 美商高通公司 | Near field user detection using a radar signal analysis associated with a user equipment |
CN111239728A (en) * | 2020-02-26 | 2020-06-05 | 深圳雷研技术有限公司 | Passenger counting method and system based on millimeter wave radar |
CN111538013A (en) * | 2020-05-12 | 2020-08-14 | 上海眼控科技股份有限公司 | Radar echo extrapolation method and device, computer equipment and storage medium |
CN112305576A (en) * | 2020-10-31 | 2021-02-02 | 中环曼普科技(南京)有限公司 | Multi-sensor fusion SLAM algorithm and system thereof |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWI800471B (en) * | 2022-11-09 | 2023-04-21 | 元智大學 | Method for counting number of people based on mmwave radar |
Also Published As
Publication number | Publication date |
---|---|
TW202318031A (en) | 2023-05-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhao et al. | mid: Tracking and identifying people with millimeter wave radar | |
JP2020071226A (en) | Fall detection method and apparatus | |
CN110361727A (en) | A kind of millimetre-wave radar multi-object tracking method | |
CN103605110B (en) | Based on the indoor passive object localization method of received signal strength | |
Zhao et al. | Human tracking and identification through a millimeter wave radar | |
US20140266860A1 (en) | Method and system for activity detection and classification | |
CN113267773B (en) | Millimeter wave radar-based accurate detection and accurate positioning method for indoor personnel | |
CN109782267B (en) | Track correlation method and vehicle-mounted radar | |
CN112929826B (en) | Indoor hybrid positioning method based on UWB triangulation positioning and fingerprint information | |
CN106468771B (en) | A kind of multi-target detection and tracking method under high clutter conditions of low Observable | |
WO2021238710A1 (en) | Method and apparatus for identifying human hand and gestures, and display device | |
TWI772208B (en) | Method for counting number of people based on mmwave radar | |
Hyun et al. | Human-vehicle classification scheme using doppler spectrum distribution based on 2D range-doppler FMCW radar | |
Arsan et al. | A Clustering‐Based Approach for Improving the Accuracy of UWB Sensor‐Based Indoor Positioning System | |
CN116087943A (en) | Indoor falling detection method and system based on millimeter wave radar | |
US20240053464A1 (en) | Radar Detection and Tracking | |
CN105388470B (en) | A kind of method of estimating target motion parameter | |
CN107801168B (en) | Outdoor self-adaptive passive target positioning method | |
CN117471453A (en) | Human motion state detection method based on millimeter wave radar | |
CN116236173B (en) | Intelligent care monitoring management system based on millimeter wave radar | |
US10942267B2 (en) | Video object processing | |
CN113820704A (en) | Method and device for detecting moving target and electronic equipment | |
CN109239677A (en) | A kind of environment self-adaption CFAR detection thresholding determines method | |
Jiang et al. | Continuous tracking of indoor human targets based on millimeter wave radar | |
CN110412531A (en) | Receiving station's method for optimizing route in the case of clutter based on amplitude information |