201118407 六、發明說明： 【發明所屬之技術領域】 本發明係關於一種室内定位方法，尤其是一種運用改 良式機率類神經網路進行定位之室内定位方法。 【先前技術】 全球定位系統(Global Positioning System，GPS)係目前 隶廣為應用的定位糸統’其主要利用三角定位的原理來進 行定位。然而GPS主要用於室外的定位，而在室内環境中 ’會受到障礙物的屏蔽效應影響，無法收到有效的衛星訊 號，因此難以應用在至内ί衣境定位中（indoor positioning)。 此外，GPS在區域較小的場合其定位精確度亦不穩定。 在室内定位中’並沒有因為環境空間的縮小而降低定 位的複雜度’反而因為諸多的因素而增加了定位的困難度 。舉例來說，室内環境的障礙物、干擾等等，都會對定位 產生干擾。 無線技術的特性非常適合應用於室内定位系統。目前 可支援室内定位的無線網路技術包含：無線區域網路 (Wireless Local Area Network，WLAN)、無線感測器網路 (Wireless Sensor Network，WSN)、無線射頻辨識系統(Radi〇 Frequency Identification，RFID)、藍芽（Bluetooth)等。其中 著名的WSN之一紫蜂(Zigbee)網路在節點與網路管理便利 、低功耗、支挺接收信號強度顯示(Received Signal Strength Indication，RSSI)功能等各種特點，使其在室内定位的角色 上受到重視。此外在位置的量測中，三角測量法、場景分 201118407 析、鄰近點等三種 別或組合使用來、棄』二可用於定位，因此常為定位系統分 力它* A V、到定位的目的。 至内疋位中，當 訊號強度(ReeeiVed . 二角量測法計算無線網路之接收 式所設計之系統二^屯如1 Stren§th，RSS) ’然而，以此方 消·)法無法計算或所，f因為RSS的訊息誤差而導致三角量 頰技術而言，欲请j二的偏差值太大。實際上以目前的射 干擾的情況下心：全正確的RSS量測值，不僅在受到 值更是無法準確’ 甚至在無干擾的空_咖的數 在來源資tK受損的心：廷端與接收端之間的距離關係。 應式的演算方式在：定;需應用-個具備容錯功能、適 ，但是°㈣經網路具備上述特性 練，不利於應用於室内的定位/疋而要大里的5十异以及訓 基於上述原因，需要—絲— 改善無線_RSS數值誤 至± (位方法’能有效的 問題。 雜㈣過大時，造成定位計算錯誤的 【發明内容】 八本發明係提供一種定位方法，其主要係當於—室内^ “夺，能提供一種精確的定位方法，為本發明之目的。每 為達到前述發明目的，本發明所運用之技術手 由該技術手段所能達到之功效包含有： 错 -種使用改良式鱗轉經鱗之定位方法， 定一待測物於—平面空間之1位座標，該平面空間传t 互相垂直之-第-軸向和-第二軸向所構成，該定位座標 201118407 係由該待測物於該第一轴向和該第二軸向之定位值表示。 該方法包含設置複數信號感測器於該平面空間；提供對應 於該第一軸向之一第一訓練資料集，其中該第一訓練資料 集包含複數第一類別向量；決定該待測物之一輸入訊號於 該第一轴向上對應於該等複數信號感測器之一第一輸出向 量集，其中該第一輸出向量集包含複數第一輸出向量丨根 據一第一公式以及一第二公式計异該待測物於該第一轴向 之定位值；提供對應於該第二軸向之一第二訓練資料集， 其中該第二訓練資料集包含複數第二類別向量；決定該待 測物之該輸入訊號於該第二轴向上對應於該等複數信號感 測器之一第二輸出向量集’其中該第二輸出向量集包含複 數第二輸出向量；及根據該第一公式以及該第二公式計算 該待測物於該第二軸向之定位值。 其中該第一公式係ΦΜ,,σ) = expf-θ~…匕，201118407 VI. Description of the Invention: [Technical Field] The present invention relates to an indoor positioning method, and more particularly to an indoor positioning method for positioning using a modified probability neural network. [Prior Art] The Global Positioning System (GPS) is currently a widely used positioning system, which uses the principle of triangulation to locate. However, GPS is mainly used for outdoor positioning, and in indoor environments, it is affected by the shielding effect of obstacles, and it cannot receive effective satellite signals, so it is difficult to apply it to indoor positioning. In addition, GPS positioning accuracy is also unstable in small areas. In indoor positioning, 'there is no reduction in the complexity of positioning due to the shrinking of the environmental space', but the difficulty of positioning is increased due to a number of factors. For example, obstacles, interference, etc. in the indoor environment can interfere with positioning. The characteristics of wireless technology are ideal for indoor positioning systems. Currently, wireless network technologies that support indoor positioning include: Wireless Local Area Network (WLAN), Wireless Sensor Network (WSN), Radio Frequency Identification System (Radi〇 Frequency Identification, RFID). ), Bluetooth, etc. One of the well-known WSN Zigbee networks is characterized by convenient node and network management, low power consumption, and Received Signal Strength Indication (RSSI) functions. The role is valued. In addition, in the measurement of position, triangulation, scene analysis 201118407 analysis, adjacent points, etc., can be used for positioning, and can be used for positioning. Therefore, it is often used for positioning system to divide it into the target. In the inner position, when the signal strength (ReeeiVed. Two-point measurement method to calculate the wireless network receiving system designed system such as 1 Stren§th, RSS) 'However, this method can not Calculate or what, f because of the RSS error message caused by the triangulation technique, the deviation value of j2 is too large. In fact, in the case of the current radio interference, the full correct RSS measurement value is not only inaccurate in the value received, even in the uninterrupted empty _ café number in the source of the damaged tK: Ting Duan and The distance relationship between the receiving ends. The calculation method of the formula is: fixed; needs to be applied - a fault-tolerant function, suitable, but ° (four) through the network with the above characteristics of training, is not conducive to the application of indoor positioning / 疋 要 大 大 大 大Reason, need - silk - improve the wireless _RSS value to ± (bit method 'can be effective. Miscellaneous (four) is too large, causing positioning calculation error [invention content] Eight inventions provide a positioning method, which is mainly In the case of indoors, it is possible to provide an accurate positioning method for the purpose of the present invention. For the purpose of the foregoing invention, the technical effects of the technical hand used by the present invention can be achieved by the technical means: Using a modified scale-turning scale positioning method, a 1st coordinate of the object to be measured is defined in the plane space, and the plane space is composed of mutually perpendicular - first-axis and - second axis, the positioning coordinates 201118407 is represented by the position of the object to be tested in the first axis and the second axis. The method includes setting a complex signal sensor in the plane space; providing one corresponding to the first axis a training data set, wherein the first training data set includes a plurality of first category vectors; determining that one of the input signals of the object to be tested corresponds to the first output vector set of one of the plurality of signal sensors in the first axis The first output vector set includes a plurality of first output vectors 计 a positioning value of the object to be tested in the first axis according to a first formula and a second formula; providing corresponding to the second axis a second training data set, wherein the second training data set includes a plurality of second category vectors; determining the input signal of the object to be tested in the second axis corresponds to one of the plurality of signal sensors An output vector set 'where the second output vector set includes a plurality of second output vectors; and calculating a position value of the object to be tested in the second axis according to the first formula and the second formula. ΦΜ,,σ) = expf-θ~...匕,
V 2σ- J ^Σ,γ/Φ(χ,ε,,σ) 該第二公式係(χ)=气-。 ί=1 【實施方式】 為讓本發明之上述及其他目的、特徵及優點能更明顯 易懂，下文特舉本發明之較佳實施例，並配合所附圖式， 作詳細說明如下： 本發明於X轴和y軸形成之一平面空間上設置複數信 號感測器，並以一待測物於該平面空間上到處移動來估測 201118407 該待測物之座標。該複數信號感測器可以包含藍芽 (Bluetooth)裝置以及家用無線基地台（Access p〇int，ap)， 而該待測物可以是一行動台（Mobile Station)。本發明之定 位架構係於ZigBee热線感測器網路下接收行動台之訊號 強度，並以改良式機率類神經網路(Modified Probabilistic. Neural Network，MPNN)來求得待測物之座標位置，細節 如以下詳述。 MPNN具有四層神經元組’分別為輸入訊號層（inpUtV 2σ- J ^Σ, γ/Φ(χ, ε, σ) The second formula is (χ) = gas-. BRIEF DESCRIPTION OF THE DRAWINGS The above and other objects, features and advantages of the present invention will become more <RTIgt; The invention provides a complex signal sensor in a plane space formed by the X-axis and the y-axis, and estimates the coordinates of the object to be tested in 201118407 by moving a test object around the plane space. The complex signal sensor may comprise a Bluetooth device and a home wireless base station (Access p〇int, ap), and the object to be tested may be a mobile station. The positioning architecture of the present invention receives the signal strength of the mobile station under the ZigBee hot line sensor network, and obtains the coordinate position of the object to be tested by using a Modified Probabilistic Neural Network (MPNN). The details are as detailed below. MPNN has four layers of neuron groups' respectively as input signal layers (inpUt
Layer)、類別層（Pattern Layer)、總合層（Summing Layer)與 輸出層（Output Layer)。假設MPNN具有類別向量c之訓練 資料集，如下所示： C={C,, C2,..·, Cm} (1) 其中，m為類別c的數量。在此情況下，對於輸入訊 號X而言，其最接近類別ci有一對應之輸出向量yi ,表示 如下： y={y]，y],···，）、} (2) 其中，m為類別c的數量。 在MPNN中，其提供一機率密度函數，表示如不： (3)Layer), the Pattern Layer, the Summing Layer, and the Output Layer. Assume that the MPNN has a training data set of the category vector c as follows: C = {C,, C2, .., Cm} (1) where m is the number of categories c. In this case, for the input signal X, its closest to the category ci has a corresponding output vector yi, which is expressed as follows: y={y], y],···,), } (2) where m is The number of categories c. In MPNN, it provides a probability density function, which means no if: (3)
/ Φ(χ.ς.σ) = exp - V 2σ2 , 其中，σ為高斯函數之平滑係數。 201118407 根據上述的公式，絕對輪出)v(x)可表示為： 1)1 跑=气--(4) Σ-7,φ(χ，〜σ) /=1 其中a為類別向量Ci内樣本之數量。此外，』可為〗和 2，分別代表待定位點之x軸與y軸座標。換言之，當卜】 時，公式(4)的女⑻可得到待定位點之χ軸座標，而當 時’公式(4)的j)2(X)可得到待定位點之y軸座標。/ Φ(χ.ς.σ) = exp - V 2σ2 , where σ is the smoothing coefficient of the Gaussian function. 201118407 According to the above formula, the absolute rotation) v(x) can be expressed as: 1) 1 run = gas - (4) Σ -7, φ (χ, ~ σ) / = 1 where a is the category vector Ci The number of samples. In addition, 』 can be 〗 〖 and 2, respectively representing the x-axis and y-axis coordinates of the point to be located. In other words, when 卜], the female (8) of formula (4) can obtain the ordinate coordinate of the point to be located, and at the time 'j) 2(X) of formula (4) can obtain the y-axis coordinate of the point to be positioned.
芩見第1圖，其係繪示本發明之利用MPNN定位的示 意圖。第1目巾，MPNN之輸人層表示所輸人之輸入訊號 X，並將該輸入訊號x與訓練資料集c内每一類別向量… 至cm的差值求出。MPNN之類別層用以根據公式(3)計管 輸入訊號X與訓練資料4之間的相似度，並與^相乘後= 由總合層加總得到公式(9)的分母，以及與々和％相乘後經 由總合層加總得到公式(9)的分子。Referring to Figure 1, there is shown a schematic representation of the present invention utilizing MPNN positioning. In the first item, the input layer of the MPNN indicates the input signal X of the input person, and the difference between the input signal x and each category vector ... to cm in the training data set c is obtained. The class layer of the MPNN is used to calculate the similarity between the input signal X and the training data 4 according to the formula (3), and multiply by ^ = the sum of the total sum of layers to obtain the denominator of the formula (9), and After multiplying by %, the molecules of the formula (9) are obtained by summing up the total layers.
很锻以上公式⑴至(4)，當j=1時，可得到待測 於X軸座標之值，亦即其x軸之座標位置。欲求得待貝* 對應於y轴座標之值(亦即#j=2時)，只要重複執行= ，至(4)即可。但必須注意的是，當j=2時(求得待測: 應於y軸座標之值），公式⑴之訓練資料集係與求/曰 測物對應於X軸座標之值)之訓練資料集c ^ 到的輸出向量亦不同。 此戶斤 根據以上的方式，本發明可較精確地求出待 標值，本發明之實驗數據將於以下描 ΛΑ Μ便大顯本發明 201118407 相較於傳統三角定位方式之優點。 參見第2圖’其係續、示複數待測點於一平面空間之擺 放示意圖。在第2圖中’係於該平面空間内（〇,〇)、㈣、 (6,0)和(6繼四個座標所圍繞的區域中均自麟〗％個待 測點，而(〇,〇)、（0,6)、（6,0)和(6,6)四個座標上則擺放四組 基地台設備卜2、3和4。基於第2圖之待測點擺放架構， 本發明先以傳統之二角定位法對每個待測點進行定位，所 得到的定位結果可麥照第3圖。在第3圖的定位結果中， 顯示某些待測點之定位座標超出該四個座標所圍繞的區 域，所得到的結果不甚準確。接著，使用本發明之MPNN 定位法對該196個待測點進行定位，其定位結果可參見第 4圖。在第4圖中，顯示所有的待測點之定位座標係介於 該四個座標所圍繞的區域，所得到的定位結果較為集中及 準確。 本發明之疋位方法’係於ZigB ee無線感測器網路下接 收行動台之訊號強度，並以改良式機率類神經網路來求得 待測物之座標位置，以達到精確定位的功效。 雖然本發明已利用上述較佳實施例揭示，然其並非用 以限定本發明’任何熟習此技藝者在不脫離本發明之精神 和範圍之内，相對上述實施例進行各種更動與修改仍屬本 發明所保S隻之技術範®壽，因此本發明之保護範圍去視後附 之申請專利範圍所界定者為準。 田 201118407 【圖式簡單說明】 第1圖：本發明之利用改良式機率類神經網路進行定位 的示意圖。 第2圖：本發明定位實驗之複數待測物於一平面空間的 擺放示意圖。 第3圖：習知三角定位法之定位結果示意圖。 第4圖：本發明利用改良式機率類神經網路定位之定位 結果示意圖。 【主要元件符號說明】 〔本發明〕 1、2、3、4 基地台設備Very forging the above formulas (1) to (4), when j=1, the value to be measured on the X-axis coordinate, that is, the coordinate position of the x-axis, can be obtained. If you want to wait for the value of the y-axis coordinate (that is, when #j=2), just repeat =, to (4). However, it must be noted that when j=2 (required to be tested: the value of the y-axis coordinate), the training data of the formula (1) and the training data corresponding to the X-axis coordinate) The output vector from set c ^ is also different. According to the above manner, the present invention can accurately determine the value to be marked, and the experimental data of the present invention will be described below to show the advantages of the present invention 201118407 compared to the conventional triangular positioning method. See Fig. 2' for the continuation, showing the arrangement of the complex points to be measured in a plane space. In Fig. 2, 'in the plane space (〇, 〇), (4), (6, 0) and (6, in the area surrounded by the four coordinates, from the column, % of the points to be measured, and Four groups of base station equipments 2, 3 and 4 are placed on the four coordinates of 〇), (0,6), (6,0) and (6,6). Placement of the points to be measured based on Fig. 2 The invention firstly locates each point to be measured by the traditional two-corner positioning method, and the obtained positioning result can be photographed in the third picture. In the positioning result of the third figure, the positioning of some points to be measured is displayed. The coordinates are not accurate enough to exceed the area surrounded by the four coordinates. Then, the 196 points to be measured are located using the MPNN positioning method of the present invention, and the positioning result can be seen in Figure 4. In the figure, it is shown that the positioning coordinates of all the points to be measured are in the area surrounded by the four coordinates, and the obtained positioning result is more concentrated and accurate. The clamping method of the present invention is based on the ZigB ee wireless sensor network. The signal strength of the mobile station is received under the road, and the coordinate position of the object to be tested is obtained by the improved probability neural network to achieve precise positioning. Although the present invention has been disclosed in the above-described embodiments, it is not intended to limit the invention. It is to be understood by those skilled in the art that various modifications and changes may be made to the above-described embodiments without departing from the spirit and scope of the invention. It is the technical scope of the invention, and the scope of protection of the present invention is subject to the definition of the patent application scope. Field 201118407 [Simple description of the drawing] Figure 1: Improvement of the use of the present invention Schematic diagram of the positioning of the probability-like neural network. Figure 2: Schematic diagram of the placement of the plurality of analytes in a plane in the positioning experiment of the present invention. Figure 3: Schematic diagram of the positioning results of the conventional triangulation method. : The present invention utilizes a modified probability-like neural network to locate a positioning result. [Main component symbol description] [Invention] 1, 2, 3, 4 base station equipment