TWI417796B - Method of recognizing objects in images - Google Patents

Description

Method of identifying objects in images

The present invention relates to a method for recognizing an object in an image, and more particularly to a method for recognizing an object in an image that accurately and quickly recognizes the shape, size or position of the captured image.

A conventional object recognition system includes a camera (or similar device) for capturing an input image, and includes cutting an input image, extracting an object from the cut image, representing the input image object, and extracting the object to be recognized classification.

The captured input image is generally composed of color or grayscale pixel data in a digital format and arranged in a two-dimensional array of X and Y directions, wherein the image may contain thousands or tens of thousands of independent pixels.

At the lowest level of the image representation, the object can be modeled as a full image, and the image is compared to the original input image in the database, pixel by pixel. However, most object recognition systems use different cut images and methods of capturing, representing, and classifying image objects to increase speed and/or accuracy (Krumm, U.S. Patent 7,092,566).

In particular, the color strip chart has been used in a variety of different processing steps, mainly to improve the speed of object recognition in color images.

In US Patent No. 7,020,329, Prempraneerach et al. describe a technique for cutting a color image into a plurality of regions using a color strip chart by converting the image into a three-dimensional color space and generating each of the color spaces. a color strip chart of the dimension, using the strip chart to generate a plurality of connection boxes in the three-dimensional color space, to calculate a normalized variation value of each connection box, and forming a connected box group, corresponding to the same A clustered pixel image of the image region corresponding to the same color characteristic is cut to extract the cut region from the image, and the clustered pixels in the image domain are classified to identify the object in the image. For the classification method, a neural network (with adaptive template matching technology), frequency-sensitive competition learning, opponent penalty competition learning or statistical classification and classification method can be used.

Prempraneerach et al. stated that the proposed clustering technique is more effective than the past iterative clustering technique, and also pointed out that the proposed cutting method is only processed once per pixel, reducing computation time to establish The intensity, chroma, and saturation of the long chart required for subsequent clustering.

However, this method still requires a series of complicated calculations before the object recognition process is achieved. First, the input image must be processed by an edge reservation filter to smooth the color image and reduce the discontinuity in the subsequently determined color strip chart. The pixels in the image are then converted from RBG to LUV, and convert the LUV into the IHS color space value by the equation group, calculate the IHS strip chart, filter the strip chart to remove the high frequency noise, and the lowest point in each strip chart Find out to form a junction box (by means of a Gaussian kernel or a Gaussian filter to spiral each strip chart).

Next, the connected cassette group is formed by calculating the normal variation of the pixel values in each of the connection boxes, and the connection boxes are connected into a tree structure according to the normalized variation values. The minimum normalized variability of the region of the junction box is used as the root node of each tree structure, and the rest is linked to the branch node of the root node by a steepest gradient descent algorithm.

In order for this method to work properly, the homogenous regions in the image must be clustered into well-defined spacings in the three-dimensional strip chart, but the true image color variation may be difficult to convert from noise or RGB to IHS color space. The color variation phase formed by linear transformation is separated.

In U.S. Patent Nos. 7,092,566, 6,952,496 and 6,611,622, Krumm describes a technique for using color strip charts to represent and classify an input image. The object recognition method firstly establishes and stores a model strip chart of the object to be identified, and then cuts an input image to extract an area that may be corresponding to the object to be recognized, and obtains a long graph from the cut areas. Then, the long graph and the stored model strip graph are obtained. If the similarity of the two long bars exceeds a predetermined threshold, the input image is matched to the model object. The matching input bar graph can also be added to the database of the model strip chart of the fixed object.

The above method establishes a model and input image strip chart by determining the true RGB color represented by the pixel in a model or input image area, and dividing the entire range of the real pixel color into a series of discrete color ranges or quantized colors. The category assigns each pixel of the extracted model or input image area to the quantized color category and establishes a count value for the number of pixels assigned to each quantized color category. In a preferred embodiment, the RGB pixel values are quantized into 27 color categories, and the comparison of the input image and the model strip chart is performed by calculating the pixel count value in each of the quantized color categories. The model strip chart must be obtained from a prefatory image similar to the input object to be identified. The image area of the model strip chart is also used in the subsequent input image to extract the object to be identified.

In a preferred embodiment, the model image is cut by: analyzing the time series images from the same image device, and determining a static background image by distinguishing pixel values that are not significantly changed on the time series image, and extracting from a subsequent image. The background image forms a foreground image and cuts the foreground image into object regions by distinguishing each of the smoothed pixel values.

However, this method is mainly used to track objects in time-series images. In addition, the method still requires a large amount of time to establish a model strip chart and compare the processing time of the input image strip graph and the model strip graph. The color strip chart generation technique used in the method does not reserve the original geometric shape, which is a summary of the number of pixels of the fixed color image, so the shape, size and position of the real object cannot be determined. Finally, similar strip graphs of different objects may result in incorrect object recognition results.

In U.S. Patent Nos. 6,532,301 and 6,477,272, Krum describes the use of a co-occurrence strip chart to represent and confirm the location of a modeled object in a search image.

The above method first establishes a model image of the object, and then calculates a co-growth bar graph for each model image. The image of the model is created by the equiangular separation of the circumferences of the object, but the viewpoints of the different distances capture the image group of the object to be confirmed, and the calculation of the common growth bar chart is performed by confirming the image of the model. Each of the possible, and unique, unaligned pairs of pixels and the number of pairs of pixels that fall within the same color range.

Then, a search window of a predetermined size is generated from the overlay portion of the search image, and a common growth bar graph of each search window is a pixel color and a space established by using the technology and the model image co-growth bar graph. Produced from the range.

Finally, each model image co-growth bar chart is compared with each search window co-growth bar chart to evaluate its similarity. Each model image co-growth bar graph matching search window co-growth bar graph is set to possibly include the object to be identified, wherein the matching system is performed by a similar value greater than a critical value. Then, the position of the object to be recognized is determined to be located in a single search area having the largest similarity measure among all search windows that may include the object to be recognized, and may be repeated upward, downward, left or right. Moving a pixel position, then calculating the search window co-growth bar graph, and comparing the search window co-growth bar graph with each model strip graph to obtain a possibly higher similarity measure value. The system and method must have its search window size, color range, and range of distances selected at the beginning of the image search.

Krum illustrates several advantages of this approach, with particular reference to co-growth bar graphs representing an efficient way to identify objects in an image. By continuously tracking pixels with matching colors and having a fixed distance therebetween, the variable geometric information can be added to a regular color strip chart. Then, by considering the color and geometric information, the object recognition method can still work under the condition that the background is messy and the amount of occlusion and the object is buckling.

However, the establishment of a model image database required for long-status chart matching is quite time consuming, and by calculating the model image and the search image, the distance of each possible unique non-arranged pixel is measured, and the co-growth bar graph is calculated. It also consumes considerable computational costs.

In addition, the representation of the object does not include detailed geometric information other than the distance between the model and the searched image, and the information of the shape, size and position of the real object does not exist, so the final object position determination is not accurate, and subsequent The determination of the position continuously requires a large amount of calculation.

Furthermore, method parameters such as the size of the search window affect the object recognition accuracy, and the image must be resized so that the total size difference between the search and the model image is processed.

One of the object recognition systems is effective, and a well-defined application is to instantly identify traffic signs on a moving vehicle, and the traffic sign system must generally be able to quickly extract objects and properly classify objects.

In U.S. Patent No. 6,801,638, Jassen et al. describe a method and apparatus for identifying traffic signs that can assist in displaying the number to the viewer by means of memory. In the method and device, the image is captured by an image sensor and analyzed and classified by a classifier in an information processing unit. Then, a synthetic image of the traffic sign is generated, and the image is stored in a memory unit and displayed through a display unit.

The input image is first searched by color and/or spatial position information, and the area larger than the average probability determines the traffic sign object that may be included. The manner in which the object is identified in the determined region is to classify the image regions for the in-store characteristic data by hierarchically and sequentially by means of the discrete known characteristics of the traffic signs, such as to confirm the shape (circle or square) and Correction procedures for internal symbols, etc. The classifier compares the input object characteristic data with the typical characteristic data set stored in the memory unit, and the object is recognized when the comparison distance is lower than the group threshold.

However, the classifier so designed must be trained several times to handle image quality differences due to weather changes and light conditions. In addition, the classifier is also related to the correlation between the shape data of the input object having the shape stored in the memory unit. In general, classifiers associated with associations may not be accurate or slow. Relevance is related to the quality of training, viewing environment and/or shape data in storage.

Improving the shape data in storage requires a lot of training or a large storage database. Second, large storage databases require more processing time for correlation processing. For example, circular and square objects appear in different ovals and rectangles at different viewing angles, which may reduce the accuracy of object recognition.

The above method also requires searching for the area of the traffic signal associated with the color value and/or spatial position that the input image may contain.

In U.S. Patent No. 6,813,545, Stromme describes a system for alerting a driver to at least one particular traffic sign, which consists of an image unit, a database, an automatic identification unit, a dual-choice selection mechanism, and a sound. And/or a visual indicator, wherein the image unit is connected to the road in front of the vehicle, the database includes at least one pre-stored traffic symbol shape, and the automatic identification unit is used to detect and confirm the continuous image. The traffic sign, and the detection and confirmation is achieved by searching for the shape in the database. The dual signal selection mechanism is included in the same image to determine the distance between the vehicle and the sign. The sound and/or visual indicator informs that a confirmed traffic sign has appeared on the road in front of the vehicle.

The input images are periodically captured according to the speed of the vehicle, and each input image is analyzed in a shape recognition processor to detect the shape of the traffic signal and the shape of the traffic symbol contained in the shape information.

The shape search and recognition unit in the system uses traditional image processing methods, such as Canny edge detection, and then performs a simple pixel-by-pixel matching procedure on the processed image. The results of viewing in several directions of the shape of the logo are stored in the shape matching database. The triangular, circular or rectangular symbols in the symbol are also confirmed using a pattern or a recognition algorithm for the detected shape. Color detection is also performed to verify that the detected shape is a traffic sign.

However, edge detection and pixel-by-pixel matching methods are generally inaccurate and relatively slow. In addition, the shape and size of the same number and symbol are quite different depending on the position of the vehicle, which makes the accuracy of shape matching and object recognition performed by edge detection and pixel-by-pixel matching.

In US Patent No. 5,926,564, Masayuki Kimura describes an image recognition in a long chart mode. However, it mainly scans the X and Y axes and performs image ratios in "0" and "1" modes. Yes, in such a comparison mode, once the image capturing device is in a different angle from the subject, the shape and size of the object cannot be correctly recognized.

It can be seen that there are still many shortcomings in the above-mentioned methods of use, which is not a good design, but needs to be improved.

In view of the shortcomings derived from the above-mentioned conventional methods, the inventor of the present invention has improved and innovated, and after years of painstaking research, he finally successfully developed a method for identifying objects in the image.

The object of the present invention is to provide a minimum specific point and a maximum specific point obtained by the X-axis or Y-axis bar graph, and the polynomial regression analysis can quickly and easily determine the shape, size and position of the object. A method of identifying objects.

A second object of the present invention is to provide a method for recognizing an object in an image, which can capture different recognizable image features, such as RGB color features or grayscale image features or video image features, and quickly and accurately The image is identified by its shape, size and position so that it can be applied to traffic signs or other fields.

A method for recognizing an object in an image to achieve the above object is as follows:

Step 1: First scanning in the image data to obtain a digital input image;

Step 2: capturing one of the recognizable features of the digital input image; for example, a color feature (such as RGB or IHS color) or a black and white feature or spectral feature;

Step 3: Simultaneously establish one of the identifiable image features, the X-axis or the Y-axis long bar graph, and the X-axis or Y-axis bar graph retains the geometric information of the extracted image;

Step 4: Find the minimum specific point and the maximum specific point on the X-axis or Y-axis strip chart to find the objects in the strip chart;

Step 5: Determine the shape of the object by polynomial regression analysis by using the minimum specific point and the maximum specific point on the X-axis or Y-axis bar graph; if the object is determined to be linear, it can be judged as a triangle; The straight line is judged as a square line; if it conforms to the quadratic equation, it is judged to be a circle;

Step 6: determining the size of the object by the distance between the two minimum specific points on the X-axis or Y-axis bar graph;

Step 7: determining the position of the object by a minimum specific point on the X-axis or Y-axis bar graph;

Step 8: Grab the image of the object of other identifiable features and repeat steps 2 through 7 to identify other features in the image.

Please refer to FIG. 1 , which is a flow chart of a method for identifying an object in an image according to the present invention, which mainly includes:

Step 1: The digital image capturing device performs digital image capturing on the object to obtain a digital image data. In addition, the digital image data can also be obtained by an analog image capturing device, and then the analog signal is converted. The digital signal can also take a digital image, and the digital image data is the object to be identified; further, the digital image capturing device or the analog image capturing device can be a camera or a camera or other device capable of capturing images. ;

Step 2: The digital image can be identified and extracted 802; for example, when the image is a color image, the identifiable image feature of a single color can be captured according to the color characteristics of RGB; or, when the image is gray In the case of the order image, the image can be captured to identify the features of the gray scale; or, when the digital image data that can be captured is in the visible spectral frequency range, only the image identifiable feature of the single spectrum is captured;

Step 3: After extracting the image recognizable feature, the X-axis direction (horizontal) or Y-axis direction (vertical) strip chart 803 of the captured image can be simultaneously calculated; and the X-axis or Y-axis strip is obtained. The method of calculating the graph is to calculate the pixels on each row or column of the image recognizable feature, so that the X-axis or Y-axis strip chart can be obtained;

Step 4: After obtaining the X-axis or Y-axis bar graph, the minimum specific point and the maximum specific point of the X-axis or Y-axis bar graph are obtained, and the minimum specific point and the largest specific point are the identifiable feature images. The zero point pixel and the maximum point point pixel 804; in addition, the minimum and maximum specific point finding method is obtained by the linear search mode in the X-axis or Y-axis strip chart, and simultaneously note and record the The minimum specific point and the position of the largest specific point; in addition, different search algorithms or methods may be used to find the minimum specific point and the maximum specific point in the X-axis or Y-axis long chart data;

Step 5: Through the polynomial regression analysis method, the object shape 805 is defined according to the minimum specific point and the largest specific point obtained; for example, if the minimum specific point to the maximum specific point is linear, the object to be identified can be determined. It should be a triangle; if the minimum specific point to the maximum specific point is a straight line, it can be determined that the object to be identified should be square or rectangular; if the minimum specific point to the maximum specific point is in accordance with the quadratic equation, it can be determined The object to be identified should be circular or elliptical;

Step 6: According to the X-axis or Y-axis long chart of the image, the position of the data and the minimum specific point and the maximum specific point can be obtained, and the pixel value of the image in the X-axis or the Y-axis direction can be obtained, and the pixel value can be obtained through the pixel value. Accurately calculate the size and size of the object 806;

Step 7: According to the X-axis and Y-axis long chart data of the image and the position of the smallest feature, the correct position of the object is determined 807;

Step 8: Retrieve the image data of different identification features (such as other characteristics of different colors or spectral frequencies) in the above digital image data, and repeat the processing flow of steps 2 to 7, so that different identification features in the same image are obtained. Perform multiple identification processes to achieve more accurate and faster identification of the shape, size and position of the object.

Please refer to FIG. 2 to FIG. 5 , which are schematic diagrams of the first implementation of the present invention. The image recognition steps are as follows:

Step 1: As shown in FIG. 2, the white background has five kinds of digital image data, which are a red triangle 1, a red circle 2, a red square 3, a green triangle 4, and a blue color. Square 5; the image can be drawn through a digital computer or captured by a digital or analog image capture device;

Step 2: As shown in Figure 3, after obtaining the digital image material, it is possible to determine the image with the red feature according to the identifiable features of the image. Therefore, the red triangle with the red feature in Figure 2 is a red circle. Both Shape 2 and Red Square 3 will be removed and image recognition will begin, instead of red features, such as Green Triangle 4 and Blue Square 5 will be removed during image capture;

Step 3: Calculate the X-axis and Y-axis strip charts; as shown in Figure 4, simultaneously calculate the red (non-black) pixels on each line in the captured image in Figure 3 to obtain the X-axis (horizontal) length. For the chart, the X-axis strip chart will display the triangle strip chart 101 corresponding to the red triangle 1, the red circle 2 and the red square 3, the curve strip chart 201 and the square strip statistics. Figure 301, and the value in the horizontal direction of the X-axis bar graph represents the number of rows, while the vertical direction represents the pixel value; the data displayed in the X-axis bar graph retains enough information to determine the original input. The shape of the object in the image and the size and position of the horizontal object; as shown in Figure 5, calculate the red (non-black) pixels in each column of the captured image in Figure 3 to obtain the Y-axis (vertical) strip statistics. Figure; the Y-axis strip chart will display the triangular strip chart 102 corresponding to the red triangle 1, the red circle 2 and the red square 3, the curve strip chart 202 and the square strip chart 302 , and the values in the horizontal direction of the Y-axis bar graph represent the number of columns. The vertical direction represents pixel values; the Y-axis strip chart data retain enough information to determine the object size and the shape of the vertical position of the object in the original input image;

Step 4: According to the X-axis or Y-axis bar graph, the minimum specific point and the maximum specific point are obtained. As shown in FIG. 4, the second minimum specific point and the largest specific point are obtained by linear search, and will be sought The minimum specific point and the position of the largest specific point (the number of rows) are recorded and recorded. Taking the X-axis long chart as an example, the pixel values of the two minimum specific points are all 0 points, so The position of the number of rows is searched line by line, and the first smallest specific point A of the triangular strip chart 101 can be searched in the 50th line, and the 50th line is noted and recorded, and then the triangle long type is searched. The second minimum specific point C of the statistical map is the 170th line, and the maximum specific point B is the pixel 85 points, and the specific points B and C searched for are recorded and recorded; then the curve strip chart 201 and the square are searched. The minimum specific points D, F, G, J and the maximum specific points E, H, and I of the long chart 301; wherein the minimum specific point D is 273 lines, F is 362 lines, G is 498 lines, and J is 563. Line, and the maximum specific point E pixel is 86 points, H and I pixels are 86 points; finally search to the end of the X-axis strip chart Similarly, the minimum specific point and the maximum specific point of the triangular strip chart 102, the curve strip chart 202 and the square strip chart 302 shown in the Y-axis bar graph shown in FIG. 5 are also in the same manner. The minimum specific points K and M of the triangular strip chart 102 are 50 rows and 136 rows, respectively, and the maximum specific point L is 119 dots; the minimum of the curve bar graph 202 is obtained. The specific points N and P are 160 lines and 245 lines, respectively, and the highest specific point O is 88 points; the minimum specific points Q and T of the square strip chart 302 are 280 lines and 365 lines, respectively, and the highest specific points R and S For the pixel 65 points, sequentially search for the terminal of the Y-axis long chart;

Step 5: After obtaining the minimum specific point and the maximum specific point, please refer to Figure 4, and perform regression analysis through the X-axis bar graph to determine the shape of the possible object; For the minimum specific point and the position of the largest specific point, the lines 1011 and 1012 of the triangular strip chart 101 and the line 2011 of the curve strip chart 201 and the lines 3011, 3012, 3013 of the square strip chart 301 are defined. Boundary shape

First, the results of the polynomial regression analysis for the line 1011 of the triangular strip chart 101:

Degree 1:-70.28+1.398x, α=0.05, p<0.0001, p<0.0001, R^2=1.00;

Degree 2: -70.71+1.41x-7.0424e-5x^2, a=0.05, p<0.0001, p<0.0001, p=0.6085, R^2=1.00;

The results show that the triangular strip chart 101 is likely to be linear with the slope 1.398, and the results of the multiple regression analysis show that the second level (Degree 2) is statistically insignificant at a=0.05.

The result of the polynomial regression analysis for line 1012 is as follows:

Degree 1:237.2-1.398x, a=0.05, p<0.0001, p<0.0001, R^2=1.00

Degree 2:238.5-1.418x+7.0424e-5x^2, a=0.05, p<0.0001, p<0.0001, p=0.6085, R^2=1.00

The results show that the triangular strip chart 101 may be linear with the slope 1.398, and the results of the polynomial regression analysis show that the second level item (Degree 2) is statistically insignificant when α=0.05.

Therefore, the joint result of the lines 1011 and 1021 shows that the object is a red triangle 1 in Fig. 2; then, the polynomial regression analysis of the line 2011 of the curve strip chart 201 is as follows:

Degree 1:48.71+0.05037x, α=0.05, p=0.0784, p=0.5589, R^2=0.00;

Degree 2: -3331 + 21.48x - 0.03375x^2, α = 0.05, p < 0.0001, p < 0.0001, p < 0.0001, R^2 = 0.95;

The results show that the results of the curve bar graph 201 are likely to be round or oval, and the results of the polynomial regression analysis show that the second term (Degree 2) is statistically significant at α=0.05.

Therefore, the result shows that the object in Figure 2 is a red circle 2.

Then, since the lines 3011, 3012, and 3013 of the square strip chart 301 are vertical and horizontal lines, the polynomial regression analysis cannot be performed; however, the line 3011 represents a line from the minimum specific value of the zero point to the maximum specific value, and the line 3012 contains all equal maximum specific values, and line 3013 represents a straight line from the maximum specific value to the minimum specific value of zero value, so the constituent objects of the lines 3011, 3012, and 3013 are a red square 3.

Step 6: After knowing the shape of the object, the object size is determined. As shown in FIG. 4, the minimum specific points A and C of the triangular strip chart 101 are 50 lines and 170 lines, and the maximum specific point B is 85. The pixels, therefore, the actual size of the red triangle 1 can be determined by the camera and image characteristics and the position of the camera in the actual environment.

The minimum specific points D and F of the curve bar graph 201 are 273 lines and 362 lines, and the maximum specific point E is 86 pixels. Therefore, if the size of the red circle 2 of FIG. 2 is to be calculated, the camera and image characteristics can be obtained. Determined with the position of the camera in the actual environment.

The minimum specific points G and J of the square strip chart 301 are 498 lines and 563 lines, and the maximum specific points H and I are 86 pixels. Therefore, if the size of the red square 3 in FIG. 2 is to be calculated, the camera and The image characteristics are determined by the position of the camera in the actual environment.

Step 7: Finally, calculate the position of the object. Please refer to Figure 4 and Figure 5 below. The minimum specific points A, C, D, F, G obtained through the X-axis strip chart and the Y-axis strip chart. J, K, M, N, P, Q, T determine the position of the object, wherein the minimum specific points A and C of the X-axis of the triangular strip chart 101 are 50 to 170; and the minimum specific point K of the Y-axis And M is 52 to 136. The position of the triangular stripe chart can be known through the minimum specific points A, C, K, and M of the X-axis and the Y-axis; the X-axis of the curve strip chart 201 is the smallest. The specific points D and F are 273 to 362; and the minimum specific points N and P of the Y axis are 160 to 245. The minimum specific points D, F, N, and P of the X and Y axes can be used to know the long curve. The position of the bar chart; the minimum specific point G and J of the X-axis of the square bar chart 301 is 498 to 563; and the minimum specific point Q and T of the Y-axis is 280 to 365, and the minimum of the X-axis and the Y-axis At a specific point G, J, Q, T data, you can know the position of the square strip chart; in addition, when necessary, the size of the real object can usually be determined by the camera and image characteristics and the camera in the actual environment. The position is judged.

Step 8: The image data of the green identification feature or the blue identification feature can be exchanged, and the steps of the steps 2 to 7 can be repeated to obtain the shape, size and position of the identified object.

Therefore, the image recognition system of the present invention is completed under the condition of no pattern comparison and pattern comparison database, and the object recognition method improves the object recognition speed and accuracy. In addition, the object shape information is obtained by using polynomial regression analysis, and thus the measurement of the shape identification accuracy is also provided according to the R^2 suitable analysis. In addition, there is no inaccurate and time-consuming input image search when identifying the shape of the object.

Please refer to FIG. 6 to FIG. 9 at the same time, which is a schematic diagram of the second embodiment of the present invention, which is mainly applied to image recognition of traffic signals, and the main identification steps are as follows:

Step 1: As shown in FIG. 6, the traffic sign is taken by the image capturing device to obtain the traffic sign image 6 of the road. The traffic sign image 6 has a red outer ring 61 and a white inner ring 62. And the black font 63 and other image recognizable features, at the same time, the traffic sign image 6 also contains other red features, such as: safflower 71 on the tree and a plurality of safflower 71 in the shrub, the safflower 71 is partially covered by green leaves Obscured

Step 2: As shown in Figure 7, the image data with red features in Figure 6 is captured, so the image features of the red outer ring 61 of the traffic sign image 6, the safflower 71 on the tree, and the safflower 71 in the bush can be obtained. ;

Step 3: As shown in FIG. 8 and FIG. 9 , the X-axis and Y-axis bar graphs are obtained through calculation, and the X-axis and Y-axis bar graphs include a red outer ring strip graph 501 and a tree. The safflower chart 502 and the plurality of safflower statistics 503 on the bush, wherein the unique figure of the long chart 501 is a feature of the shape of the original ring object; and the red outer ring is known by the Y-axis bar chart of FIG. 61 overlaps with the safflower 71 on the tree in the Y-axis (vertical) direction, so the red outer ring strip graph 501 and the tree safflower strip graph 502 data are added to each other to match the red color corresponding to the object to be identified. The ring strip chart 501 adds noise to the data; in addition, the unique graph of the long chart 501 is the feature of the original ring shape; the calculation mode of the X-axis and Y-axis strip charts and Figure 4 and The same is true, and it is not mentioned here;

Step 4: As shown in FIG. 8 and FIG. 9 , the X-axis and Y-axis long bar graph data are scanned from zero rows to 3264 rows to obtain the minimum specific point A of the X-axis red outer loop strip graph 501. ', D' and the maximum specific point B', C' and the Y-axis red outer ring strip chart 501 minimum specific points E', H' and the maximum specific points F', G'; the minimum specific point A' is 1112 lines, D' is 2040 lines, E' is 1099 lines, H' is 2078 lines; the maximum specific points B' and C' have pixel values of 710 points, and F' and G' have pixel values of 660 and 650; In addition, the search methods of the minimum specific points A', D', E', H' and the maximum specific points B', C', F', G' are the same as those in FIG. 4 and FIG. 5, and are not described herein. ;

Step 5: After obtaining the minimum specific point and the maximum specific point, use the three sets of lines 5011, 5012, and 5013 shown in FIG. 8 to determine the shape of the object by using polynomial regression analysis (regression);

The polynomial regression analysis of line 5011 in Figure 8 is as follows:

Degree 1:-4439+4.145x, α=0.05, p<0.0001, p<0.0001, R^2=0.93

Degree 2:-42765+68.91x-0.02733x^2, α=0.05, p<0.0001, p<0.0001, p=0.0001, R^2=0.99

The results show that the line 5011 is likely to be round or oval, and the polynomial regression analysis shows that the second level term is statistically significant at α=0.05;

The polynomial regression analysis of line 5012 in Figure 8 is as follows:

Degree 1:352.3+0.008745x, α=0.05, p<0.0001, p=0.6111, R^2=0.00

Degree 2: 6456-7.861x+0.002503x^2, α=0.05, p<0.0001, p<0.0001, p<0.0001, R^2=0.89

The results show that the line 5012 may be round or oval, and the results of multiple regression analysis show that the second level term is statistically significant at α=0.05; the results of the line 5013 polynomial regression analysis in Figure 8 are as follows:

Degree 1:9155-4.43x, α=0.05, p<0.0001, p<0.0001, R^2=0.92

Degree 2:-102241+109x-0.02887x^2, α=0.05, p<0.0001, p<0.0001, p=0.0001, R^2=0.99

The results show that line 5013 may be circular or oval, and polynomial regression analysis, the second level term is statistically significant at α = 0.05; therefore, the combined results of lines 5011, 5012, 5013 show that Figure 8 shows The strip chart 501 is likely to be circular or oval in shape.

Step 6: After the shape of the object is known, the object size is determined. Referring to FIG. 8, the minimum specific points A' and D' of the red outer ring strip chart 501 in the X-axis direction are 1112 lines and In 2040 lines, the maximum specific points B' and C' are 710 pixels; and the minimum specific points E' and H' in the Y-axis direction shown in FIG. 9 are 1099 lines and 2078 lines, and the maximum specific point F is 660 pixels. G' is 650 pixels. According to the above conditions, the correct size of the object can be calculated. In addition, the correct size of the object can be determined by the camera and image characteristics and the position of the camera in the actual environment, and will not be described here.

Step 7: Finally, calculate the position of the object. Please refer to Figure 8 and Figure 9. The minimum specific points A', D', E' obtained through the X-axis strip chart and the Y-axis strip chart. And H' judge the position of the object, and the manner of judgment is the same as that disclosed in the above FIG. 4 and FIG. 5, and is not described herein;

Step 8: Re-take the other image identifiable features and repeat steps 2~7 to re-identify the image.

Therefore, the information such as the color, shape, size and/or position of the object obtained by the method of the present invention can be applied to traffic signs and can be applied to other purposes or fields.

In addition, the safflower chart 502 on the tree shown in FIG. 8 shows that the object does not have a well-defined linearity or curve via polynomial regression analysis because the R^2 values of the first and second levels of analysis are small:

Degree 1:-1653+0.5543x, α=0.05, p<0.0001, p<0.0001, R^2=0.48

Degree 2:-155477+102x-0.01672x^2, α=0.05, p<0.0001, p<0.0001, p=0.0001, R^2=0.66

Therefore, the object is clearly not a circular or linear traffic sign object.

When using the polynomial regression to analyze the image shape, it is not affected by noise. As shown in Figure 9, the polynomial regression analysis results for the line 5014 displayed on the Y-axis strip chart are as follows:

Degree 1:393-0.02945x, α=0.05, p<0.0001, p=0.0370, R^2=0.01

Degree 2: 5350-6.353x+0.001987x^2, α=0.05, p<0.0001, p<0.0001, p<0.0001, R^2=0.9

The results show that the line 5014 of the long chart 501 may be circular or oval (not linear), and the result of the polynomial regression analysis shows that the second level term is statistically significant at α=0.05, which is longer than the X-axis of Figure 8. The line 5012 of the chart is similar, so the shape recognition of the Y-axis strip chart is not affected by the noise of the safflower on the tree.

The method for recognizing an object in an image provided by the present invention has the following advantages when compared with other conventional technologies:

1. The present invention is capable of quickly and easily determining the shape, size and position of an object in an image by using a minimum specific point and a maximum specific point obtained on an X-axis or Y-axis strip chart. Methods.

2. The invention can capture different identifiable image features, such as RGB color features or grayscale image features or video image features, and quickly and accurately identify the shape, size and position of the captured image, so that Used in traffic signs or other fields.

The detailed description of the preferred embodiments of the present invention is intended to be limited to the scope of the invention, and is not intended to limit the scope of the invention. The patent scope of this case.

To sum up, this case is not only innovative in terms of technical thinking, but also able to enhance the above-mentioned multiple functions compared with conventional articles. It should fully comply with the statutory invention patent requirements of novelty and progressiveness, and apply in accordance with the law. I urge you to approve this article. Invention patent application, in order to invent invention, to the sense of virtue.

1. . . Red triangle

2. . . Red circle

3. . . Red square

4. . . Green triangle

5. . . Blue square

6. . . Traffic signs

61. . . Red outer ring

62. . . White inner ring

63. . . Black font

71. . . safflower

101. . . Triangle strip chart

1011. . . line

1012. . . line

102. . . Triangle strip chart

201. . . Curve strip chart

2011. . . line

202. . . Curve strip chart

301. . . Square strip chart

3011. . . line

3012. . . line

3013. . . line

302. . . Square strip chart

501. . . Outer ring strip chart

5011. . . line

5012. . . line

5013. . . line

5014. . . line

502. . . Tree safflower strip chart

503. . . Shrub red flower strip chart

A, C, D, F, G, J. . . Minimum specific point

B, E, H, I. . . Maximum specific point

R, M, N, P, Q, T. . . Minimum specific point

L, O, R, S. . . Maximum specific point

A', D', E', H'. . . Minimum specific point

B', C', F', G'. . . Maximum specific point

1 is a step diagram of a method for identifying an object in an image according to the present invention;

2 is a color diagram of five objects of the first implementation of the method for identifying an object in an image;

Figure 3 is a schematic diagram of the features of the identifiable image captured in Figure 2;

Figure 4 is a graph of the X-axis color strips obtained by calculating the color pixels of each row in Figure 3;

Figure 5 is a statistical diagram of the Y-axis color strip obtained by calculating each column of color pixels in Figure 3.

Figure 6 is a second embodiment of a color traffic locomogram of the method for identifying an object in an image of the present invention;

Figure 7 is a schematic diagram of capturing the identifiable image features in Figure 6.

Figure 8 is a graph of the X-axis color strips obtained by calculating the color pixels of each row in Figure 6;

Figure 9 is a statistical diagram of the Y-axis color strip obtained by calculating the color pixels of each column in Figure 6.

No component representation symbol

Claims (20)

A method for recognizing an object in an image, comprising: step 1: first obtaining a digital image data; step 2: extracting at least one object according to the identifiable feature of the digital image; and step 3: simultaneously calculating step 2 Take the X-axis or Y-axis long chart of the image; Step 4: Obtain the X-axis or Y-axis long bar graph, and then find the minimum specific point and the maximum specific point of the X-axis or Y-axis bar graph; 5: Through the polynomial regression analysis method, the shape of each object is determined according to the minimum specific point and the maximum specific point of the identifiable object. The method for recognizing an object in an image as described in claim 1 further includes a step 6 of capturing the image data of the same image but different identifiable features in step 1, and repeating step 2 Go to the processing flow of step 5. The method for recognizing an object in an image as described in claim 1 further includes a step 6, wherein the step 6 determines the size of the object according to the minimum specific point and the position of the largest specific point. The method for recognizing an object in an image as described in claim 1 further includes a step 6 of determining the correct position of the object according to the minimum specific point and the position of the largest specific point. The method for recognizing an object in an image according to the first aspect of the patent application, wherein the digital image data obtained in the step 1 is obtained by photographing an object by a digital image capturing device. The party that identifies the object in the image as described in item 5 of the patent application scope The method wherein the digital image capturing device is a camera or a camera. The method for recognizing an object in an image according to the first aspect of the patent application, wherein the digital image data obtained in the step 1 is obtained by capturing an object by an analog image capturing device, and converting the analog signal into a digital signal. To obtain a digital image. A method for recognizing an object in an image as described in claim 7 wherein the analog image capturing device is a camera or a camera. A method for recognizing an object in an image as described in claim 1 wherein the step 2 identifies an RGB or IHS color identifiable feature characterized by a single color. The method for recognizing an object in an image according to the first aspect of the patent application, wherein the step 2 identifiable feature is a gray-scale identifiable feature. A method for recognizing an object in an image as described in claim 1, wherein the step 2 identifiable feature is a video spectral frequency identifiable feature. The method for recognizing an object in an image as described in claim 1, wherein the calculation method of the X-axis or Y-axis bar graph of the step 3 is performed on each row or column of the image recognizable feature. Pixels. The method for recognizing an object in an image as described in claim 1, wherein the minimum specific point of the step 4 is obtained on the X-axis or Y-axis bar graph, and is obtained in a linear search mode. Zero pixels. The method for recognizing an object in an image as described in claim 1, wherein the maximum specific point of the step 4 is obtained on the X-axis or Y-axis bar graph, and is obtained in a linear search mode. Maximum Value point pixels. The method for recognizing an object in an image as described in claim 1, wherein the minimum specific point and the maximum specific point position obtained in the step 4 must be noted and recorded. The method for recognizing an object in an image according to the first aspect of the patent application, wherein the polynomial regression analysis method of the step 5 is based on analyzing the line on the X-axis or the Y-axis strip chart, when the X-axis is Or the Y-axis strip chart has two lines connecting the two minimum specific points and one maximum specific point, and the two lines are linearly oriented, and it can be determined that the object to be recognized should be a triangle. For example, in the method of identifying an object in an image as described in claim 15, when there are four lines on the X-axis or Y-axis strip chart, the two minimum specific points and the two largest specific points are connected, and the four lines are in a straight line. , it can be determined that the object to be recognized should be square or rectangular. The method for recognizing an object in an image as described in claim 15 wherein the minimum specific point to the maximum specific point of the line conforms to a quadratic equation, and the object to be identified should be determined to be circular or elliptical. . A method for recognizing an object in an image, comprising: step 1: first obtaining a digital image data; step 2: extracting at least one object according to the identifiable feature of the digital image; and step 3: simultaneously calculating step 2 Take the X-axis or Y-axis long chart of the image; Step 4: Obtain the X-axis or Y-axis long bar graph, and then find the minimum specific point and the maximum specific point of the X-axis or Y-axis bar graph; Step 5: Determine the shape of each object according to the minimum specific point and the maximum specific point of the identifiable object through the polynomial regression analysis method. Step 6: Determine the size and size of the object according to the minimum specific point and the position of the largest specific point. . A method for recognizing an object in an image, comprising: step 1: first obtaining a digital image data; step 2: extracting at least one object according to the identifiable feature of the digital image; and step 3: simultaneously calculating step 2 Take the X-axis or Y-axis long chart of the image; Step 4: Obtain the X-axis or Y-axis long bar graph, and then find the minimum specific point and the maximum specific point of the X-axis or Y-axis bar graph; 5: Further, through the polynomial regression analysis method, the shape of each object is determined according to the minimum specific point and the maximum specific point of the identifiable object; Step 6: determining the size and size of the object according to the minimum specific point and the position of the largest specific point; Step 7: Determine the correct position of the object based on the minimum specific point and the position of the largest specific point.