License plate detection device and license plate detection method

Abstract

To provide a license plate detection device and a license plate detection method which prevent detection delays and reduce a computational load.SOLUTION: A license plate detection device 10 comprises a license plate detection module 11, a marginal region determination module 12, and an erroneous detection restriction module 13. The license plate detection module 11 calculates a position and likelihood of a license plate candidate. The marginal region determination module 12 determines a size of a marginal region assumed as a vehicle body. The erroneous detection restriction module 13 removes results considered as erroneous detection from detection results received from the license plate detection module 11. When multiple candidate license plates share an identical marginal region, the license plate detection device leaves a candidate with the highest likelihood and remove others so that erroneous detection can be suppressed.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an image processing device and an image processing method for detecting a license plate of a vehicle. [Background technology]

[0002] There is a known technology to detect the position of a vehicle license plate from an arbitrary image (Figure 4). This is realized by image recognition technology such as procedural algorithms and machine learning. For example, Non-Patent Document 1 detects the license plate position based on the brightness change in the image. In addition, Non-Patent Document 2 determines the position of the license plate using a convolutional neural network, which is a type of machine learning.

[0003] Furthermore, in Patent Document 1, multiple consecutive images are prepared and license plate candidate areas are predicted to prevent false detection. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Takehiko Kato, Kayo Tanaka, Masatoshi Asada, Hiroyuki Yasuhara, Kanae Tachiki, Michihiro Kodama, Yusuke Yasuhara, Toshihiro Asai, Yasuo Ogiuchi. Development of an automatic license plate reader. [D] Industrial Applications Division, Transportation and Electric Railway Research Group, 2013. [Non-Patent Document 2] Chien-Chang Chen, Yu-Yang Lin, and Jing-Chung Shen. License plate recognition model for tilt correction based on convolutional neural network. pp. 336-340, 2022. [Patent documents]

[0005] [Patent Document 1] Patent No. 6720694 Summary of the Invention [Problem to be solved by the invention]

[0006] In image recognition, it is difficult to avoid false detection of objects similar to the desired object, not only in the case of license plates. For example, in the case of license plate detection, other parts of the vehicle body or the background may be falsely detected as the license plate. In Patent Document 3, false detection is suppressed by preparing multiple consecutive images and predicting the license plate candidate area, but the method of using multiple images has problems such as delays and a large amount of calculation. In a system that requires real-time processing, a faster and more convenient method is required.

[0007] An object of the present application is to provide a license plate detection device and a license plate detection method that can reduce the delay and amount of calculations described above. [Means for solving the problem]

[0008] In order to address the above-mentioned problems, the present invention provides the following technical solutions. That is, the invention described in claim 1 is a device for detecting a vehicle license plate from a captured image, comprising: a license plate detection module that detects the position of the license plate candidate from the image and calculates the probability of whether that position is the license plate; a margin area determination module that determines the size of a margin area that assumes the apparent size of the vehicle in the image; and a false detection suppression module that expands the margin area determined by the margin area determination module for each position of the license plate candidate based on the detection results received from the license plate detection module, and, if the margin areas overlap, judges the license plate candidate with the lower probability to be a false detection and eliminates that candidate.

[0009] The invention described in claim 2 is the license plate detection device according to claim 1, characterized in that the license plate detection device detects the license plate from a single captured image.

[0010] The invention described in claim 3 is the license plate detection device described in claim 1, characterized in that the method of calculating the position of the license plate candidate and the probability that the position is the license plate is a procedural algorithm.

[0011] In addition, the invention described in claim 4 is the license plate detection device described in claim 1, characterized in that the method of calculating the position of the license plate candidate and the probability that the position is the license plate by the license plate detection module is machine learning.

[0012] The invention described in claim 5 is a method for detecting a vehicle license plate from a single captured image, which detects candidates for the location of the license plate from the single captured image and calculates the probability of whether that location is the license plate. If multiple candidates for the location of the license plate are detected, a margin area that assumes the apparent size of the vehicle captured in the image is expanded for each candidate, starting with the candidate with the highest probability, and comparing whether the expanded margin area overlaps with other margin areas. If so, the previously expanded margin area with the high probability is retained, and the other margin areas with the low probability are judged to be erroneous detections and eliminated, and the license plate detection method is characterized in that the margin areas of all candidates are expanded and the comparison is completed, thereby eliminating erroneous detections and completing the detection of the license plate. Effect of the Invention

[0013] The license plate detection device and license plate detection method of the present invention can reduce delays and the amount of calculations. In particular, when detecting a license plate from a single captured image, delays and the amount of calculations can be reduced because only a single image needs to be considered, and erroneous detection of license plates can be suppressed in a simpler manner than conventional methods. [Brief description of the drawings]

[0014] [Figure 1] 1 is a schematic diagram of the configuration of a license plate detection device of the present invention. [Diagram 2] 4 is a flowchart of the license plate detection of the present invention. [Diagram 3] 4 is an example of the operation of the license plate detection device of the present invention. [Figure 4] FIG. 1 is an image diagram of conventional number plate detection. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] The license plate detection device according to the present invention will be described with reference to the drawings. Fig. 1 is a schematic diagram of the license plate detection device according to the present invention, Fig. 2 is a flowchart of license plate detection according to the present invention, Fig. 3 is an example of the operation of the license plate detection device according to the present invention, and Fig. 4 is an image diagram of conventional license plate detection. Note that this embodiment shows a detection device and a detection method for detecting a license plate from a single captured image.

[0016] Generally, each vehicle has one license plate installed on both the front and the back of the vehicle (some vehicles, such as motorcycles, only have one installed on the rear of the vehicle). Therefore, license plates should be spaced apart by a distance equal to the size of the vehicle. If two or more license plate candidates are detected in extremely close proximity, all but one are likely to be false positives. Therefore, when a license plate is detected, false positives can be reduced by eliminating all objects other than the one that is most likely to be a license plate within a range of vehicle size that is previously assumed.

[0017] The configuration of the device is shown in Figure 1. The license plate detection device 10 has a license plate detection module 11, a margin area determination module 12, and an erroneous detection suppression module 13.

[0018] The license plate detection module 11 calculates the position and probability of a license plate candidate. The higher the probability, the more likely the candidate is to be a license plate. Any calculation algorithm can be used as long as it can calculate the position and probability of a candidate. For example, there are methods such as template matching, which is a procedural algorithm, and convolutional neural networks, which are machine learning methods. The calculation results are sent to the false positive suppression module 13.

[0019] The margin area determination module 12 determines the size of the margin area assumed to be the vehicle body. The size of the margin area varies depending on the mode of operation. For example, when operating with a fixed surveillance camera, the apparent size of the vehicle shown in the image can be predicted in advance, so the size may be a fixed value determined in advance. On the other hand, when the apparent size of the vehicle cannot be predicted, such as when the camera or subject moves, more accurate predictions can be made by varying the size of the margin area according to the distance between the camera and the subject, the size and type of the license plate, etc. The determined size of the margin area is sent to the false detection suppression module 13.

[0020] The false positive suppression module 13 eliminates results that are considered to be false positives from among the detection results received from the license plate detection module 11. If the margin area determined by the margin area determination module 12, i.e., the area where a vehicle body is assumed to exist, overlaps with another margin area, this does not usually occur. Therefore, of the license plate candidates that overlap the margin area, one is considered to be the true detection result, and the others are considered to be false positives. Therefore, false positives can be suppressed by keeping the candidate with the highest probability and eliminating the others. Figure 2 is a flowchart summarizing the flow of the above operations.

[0021] A detailed example of the method is shown in Figure 3. First, from among the detection results (Figure 3(a)), a margin area is developed from the center of the detection results in order of the highest probability of being a license plate (Figure 3(b)). If the margin areas of different detection results overlap (Figure 3(c)), the detection result with the lower probability is rejected as a false positive. For example, in Figure 3(c), the margin areas of (1) and (3) overlap, but since (3) has a lower probability of being a license plate, it is rejected as a false positive. By repeating the above procedure for all detection results, the correct license plate detection result is obtained (Figure 3(d)). As described above, since only one image needs to be considered, delays and calculation amounts can be reduced, and false positives of license plates can be suppressed in a simpler way than before. [Industrial Applicability]

[0022] The present invention is a device capable of reducing delays and the amount of calculation required to detect the license plates of vehicles in motion, such as in a vehicle monitoring system. [Explanation of symbols]

[0023] 10 License plate detection device 11 License Plate Detection Module 12 Margin area determination module 13 False positive suppression module

Claims

1. An apparatus for detecting a license plate of a vehicle from a captured image, comprising: a license plate detection module for detecting a position of the candidate license plate from the image and calculating a probability that the position is the license plate; a margin area determination module for determining a size of a margin area that assumes an apparent size of the vehicle in the image; a false positive suppression module that develops the margin area determined by the margin area determination module for each position of the license plate candidate based on the detection result received from the license plate detection module, and when the margin areas overlap, determines that the license plate candidate with a lower probability is a false positive and eliminates the candidate; A license plate detection device comprising:

2. The license plate detection device according to claim 1, wherein the license plate detection device detects the license plate from a single captured image.

3. 2. The license plate detection device according to claim 1, wherein the method of the license plate detection module for calculating the position of the license plate candidate and the probability that the position is the license plate is a procedural algorithm.

4. The license plate detection device according to claim 1, characterized in that the method of calculating the position of the license plate candidate and the probability that the position is the license plate in the license plate detection module is machine learning.

5. A method for detecting a vehicle license plate from a single captured image, comprising: Detecting a candidate for the position of the license plate from one acquired image and calculating the probability that the position is the license plate; When a plurality of candidates for the position of the license plate are detected, a margin area is developed for each candidate, assuming an apparent size of the imaged vehicle, starting from the candidate with the highest probability; comparing whether the developed margin area overlaps with other areas, and if so, leaving the previously developed margin area with a high probability and judging the other margin area with a low probability as a false positive and eliminating it, and repeating this process; Completing the detection of the license plate by expanding the margin areas of all candidates and completing the comparison and eliminating false positives. A license plate detection method comprising:

Images