Intelligent detection and identification system and method for container number

Abstract

The present invention provides an intelligent detection and identification system and method for container numbers, which relates to the field of warehouse management technology. The system comprises: a detection device continuously outputs a shooting signal when detecting a container vehicle entering or exiting an entrance or exit of a to-be-detected area; an image acquisition device continuously acquires images of the corresponding container vehicle based on the shooting signal to obtain multiple frames of vehicle images; a server performs container number contour recognition on each frame of the vehicle image to obtain a container number text contour; the text symbols in each container number text contour are recognized to obtain a container number recognition result in each frame of the vehicle image; and the container number of the container vehicle is obtained based on the container number recognition results. The beneficial effect is that the vehicle images of container vehicles entering or exiting the warehouse gate are automatically captured to perform intelligent detection and identification of the container number, thereby eliminating errors caused by manual transcription, reducing the time for container vehicles to enter and exit the warehouse, improving the efficiency and convenience of warehouse management, and saving labor costs.

Description

Technical Field

[0001] The present invention relates to the technical field of warehouse management, and in particular to an intelligent detection and identification system and method for container numbers. Background Art

[0002] In recent years, with the continued development of the economy, the container business has grown rapidly. As a key component of warehouse management, the demand for intelligent container management has become increasingly strong. The container number, a unique identifier for vehicles entering and leaving the warehouse, uses the ISO 6346 (1995) standard. It must be recorded at every stage of the container transportation process to facilitate the identification of the container's location during circulation.

[0003] However, most current warehouse management technologies are generally outdated. Many gates still manage container transport by manually recording container numbers, leading to long queues of container vehicles at the gates. Errors caused by manual recording increase the burden of duplicate data processing. These and other issues have led to a decline in warehouse service levels, impacting not only the warehouse's own viability but also potentially causing unnecessary financial losses for businesses. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention provides an intelligent detection and identification system for container numbers, comprising:

[0005] A detection device is provided at an entrance or exit of an area to be detected, and is used to continuously output a shooting signal when a container vehicle is detected entering or exiting the entrance or exit of the area to be detected;

[0006] An image acquisition device is provided at the entrance or exit of the area to be detected, and is used to continuously acquire images of the corresponding container vehicle according to the shooting signal to obtain multiple frames of vehicle images and output them;

[0007] A server is connected to the image acquisition device, and the server includes:

[0008] A contour recognition module is used to perform container number contour recognition on each frame of the vehicle image to obtain the container number text contour in each frame of the vehicle image;

[0009] a text recognition module, connected to the contour recognition module, for respectively recognizing the text symbols in the container number text contour corresponding to each frame of the vehicle image, and obtaining a container number recognition result in each frame of the vehicle image;

[0010] A result processing module is connected to the text recognition module and is used to obtain the container number of the container vehicle according to the container number recognition results.

[0011] Preferably, the detection device includes a controller and an ultrasonic radar and a laser sensor connected to the controller;

[0012] The controller is configured to output the shooting signal indicating that a container vehicle is detected to be entering or exiting when the detection signals from the ultrasonic radar and the laser sensor are received simultaneously.

[0013] Preferably, the image acquisition device includes multiple surveillance cameras at different angles installed at the entrances and exits of the area to be detected; and the result processing module includes:

[0014] a first verification submodule, configured to perform container number verification on each of the container number recognition results, so as to add the container number recognition results that meet the preset container number verification rules into a first result set;

[0015] a second check submodule, connected to the first check submodule, configured to match the case owner code included in each case number recognition result in the first result set with at least one pre-configured standard code, and add the case number recognition result corresponding to the case owner code that matches any of the standard codes to a second result set;

[0016] a set division submodule, connected to the second verification submodule, for dividing the second result set into a plurality of subsets according to the surveillance camera corresponding to each of the box number recognition results in the second result set;

[0017] A processing submodule, connected to the set division submodule, is used to obtain the corresponding single camera recognition result associated with the surveillance camera according to the container number recognition results in each of the subsets, and when it is determined that there are at least two of the single camera recognition results associated with the surveillance cameras that are the same, the same single camera recognition result is used as the container number of the container vehicle.

[0018] Preferably, each frame of the vehicle image is associated with an image acquisition time;

[0019] The result processing module further includes a result expansion submodule, which is connected to the set partitioning submodule and the first check submodule respectively, and the result expansion submodule includes:

[0020] a sorting unit, configured to sort the box number recognition results corresponding to each frame of the vehicle image in each subset according to the order of the image acquisition time before the processing submodule processes the box number recognition results in each subset;

[0021] an expansion unit connected to the sorting unit, configured to, for the sorted subset, have the server sequentially count the number of occurrences of each character in at least two consecutive box number recognition results according to the sorting, then use the character with the largest number of occurrences as the character of the corresponding position to form a new recognition result, and save the new recognition result as the box number recognition result to expand the box number recognition result;

[0022] The first syndrome submodule and the second syndrome submodule again perform container number verification and matching on the expanded container number recognition result in sequence to update the second result set;

[0023] The set division submodule divides the updated second result set into subsets again.

[0024] Preferably, each of the box number recognition results is associated with a corresponding confidence level;

[0025] Then the processing submodule includes:

[0026] A first statistical unit is configured to, for each subset, respectively count all result values ​​of the box number recognition results in the subset as single-camera recognition results associated with the corresponding surveillance camera;

[0027] a determination unit, connected to the first statistical unit, configured to, for each of the single-camera recognition results, output the single-camera recognition result when determining that the single-camera recognition result exists in at least two of the subsets simultaneously, and generate a first signal when determining that the single-camera recognition result does not exist in at least two of the subsets simultaneously;

[0028] a second statistical unit, connected to the judgment unit, for respectively counting, in each of the subsets, a first number of the box number recognition results having the same result value as the box number recognition result associated with the highest confidence level according to the first signal, and outputting the single camera recognition result associated with the subset having the highest confidence level, the highest confidence level being greater than a preset threshold, and the largest first number;

[0029] An output unit is connected to the judgment unit and the second statistical unit respectively, and is used to use the output single camera recognition result as the container number of the container vehicle.

[0030] Preferably, the processing submodule further includes a same period verification unit, which is connected to the judgment unit, the second statistical unit and the output unit respectively, and the same period verification unit includes:

[0031] a first check subunit, configured to add all the output single-camera recognition results to a third result set, and output a second signal when it is determined that the number of elements in the third result set is greater than one, and output a third signal when the number of elements is not greater than one;

[0032] a second check subunit, connected to the first check subunit, configured to perform bit-wise comparison on each of the single-camera recognition results in the third result set in sequence according to the second signal, and output a fourth signal when it is determined that the difference in characters between the two compared single-camera recognition results is no greater than one bit, and output the third signal when the difference in characters is greater than one bit;

[0033] a third check subunit, connected to the second check subunit, configured to calculate, based on the fourth signal, the respective proportions of the two compared single-camera recognition results in all the single-camera recognition results, and, if it is determined that the two corresponding proportions are different, delete the single-camera recognition result with the lower proportion from the third result set; and, if the two proportions are the same, delete the associated single-camera recognition result with the lower confidence level from the third result set, so as to update the third result set and subsequently output the third signal;

[0034] a fourth check subunit, connected to the first check subunit, the second check subunit, and the third check subunit, respectively, and configured to output each of the single-camera recognition results in the third result set when it is determined based on the third signal that the comparison of each of the single-camera recognition results is complete;

[0035] The output unit uses each of the single-camera recognition results in the third result set as the container number of the container vehicle.

[0036] Preferably, the server further includes a data optimization module, connected to the contour recognition module and the text recognition module, respectively, for counting a second number of the container number text contours in each frame of the vehicle image, and discarding at least the container number text contour in the corresponding vehicle image when it is determined that the second number is less than a preset value, and retaining the container number text contour in the corresponding vehicle image when the second number is not less than the preset value;

[0037] The text recognition module recognizes the text symbols in the container number text outline corresponding to each frame of the retained vehicle image.

[0038] Preferably, the server also includes a status detection module, connected to the contour recognition module, for respectively counting the text contour detection coordinates of the container number text contour in each frame of the vehicle image in the image coordinate system, and obtaining the corresponding driving direction of the container vehicle based on the text contour detection coordinates corresponding to multiple consecutive frames of the vehicle image to characterize the entry and exit status of the container vehicle.

[0039] Preferably, it further includes a visualization platform connected to the server for visually displaying at least one of the vehicle image of each frame, the container number of the corresponding container vehicle, and the entry and exit status of the container vehicle.

[0040] The present invention also provides an intelligent detection and identification method for container numbers, which is applied to the above-mentioned intelligent detection and identification system. The intelligent detection and identification method includes:

[0041] Step S1, the detection device continuously outputs a shooting signal when detecting a container vehicle entering or exiting the entrance or exit of the area to be detected;

[0042] Step S2, the image acquisition device continuously acquires the corresponding image of the container vehicle according to the shooting signal to obtain multiple frames of vehicle images and sends them to the server;

[0043] Step S3, the server performs container number contour recognition on each frame of the vehicle image to obtain the container number text contour in each frame of the vehicle image;

[0044] Step S4, the server recognizes the text symbols in the container number text outline corresponding to each frame of the vehicle image, and obtains the container number recognition result in each frame of the vehicle image;

[0045] In step S5, the server obtains the container number of the container vehicle according to the container number recognition results.

[0046] The above technical solution has the following advantages or beneficial effects:

[0047] 1) By automatically capturing multiple frames of vehicle images of container vehicles entering and exiting the inspection area, and intelligently detecting and identifying container numbers based on these images, errors caused by manual transcription are eliminated, reducing the time it takes for container vehicles to enter and exit the warehouse, improving the efficiency and convenience of warehouse management while saving labor costs.

[0048] 2) In the process of intelligent detection and identification of container numbers, multiple intelligent verification methods are used to effectively improve the accuracy of container number detection and identification, thereby greatly improving the reliability of warehouse management;

[0049] 3) By providing a visual platform, it provides warehouse managers with a more efficient and convenient online management model. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a structural diagram of an intelligent container number detection and identification system in a preferred embodiment of the present invention;

[0051] Figure 2 The figure is a flow chart of a method for intelligent detection and identification of container numbers in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0052] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. The present invention is not limited to this embodiment, and other embodiments may also fall within the scope of the present invention as long as they conform to the gist of the present invention.

[0053] In a preferred embodiment of the present invention, based on the above problems existing in the prior art, an intelligent detection and identification system for container numbers is provided. Figure 1 Shown, including:

[0054] The detection device 100 is provided at an entrance or exit of an area to be detected, and is used to continuously output a shooting signal when a container vehicle is detected entering or exiting the entrance or exit of the area to be detected;

[0055] The image acquisition device 200 is provided at the entrance or exit of the area to be inspected, and is used to continuously acquire images of the corresponding container vehicle according to the shooting signal to obtain multiple frames of vehicle images and output them;

[0056] The server 300 is connected to the image acquisition device 200 and includes:

[0057] Contour recognition module 1, used to perform container number contour recognition on each frame of vehicle image to obtain the container number text contour in each frame of vehicle image;

[0058] The text recognition module 2 is connected to the contour recognition module 1 and is used to recognize the text symbols in the container number text contour corresponding to each frame of the vehicle image, and obtain the container number recognition result in each frame of the vehicle image;

[0059] The result processing module 3 is connected to the text recognition module 2 and is used to obtain the container number of the container vehicle according to the recognition results of each container number.

[0060] Specifically, in this embodiment, by setting up a detection device 100 at the entrance and exit of the area to be detected, it is possible to detect in real time whether there are container vehicles entering or exiting the entrance and exit of the area to be detected, and when it is detected that there are container vehicles entering or exiting, it triggers the acquisition of continuous multiple frames of vehicle images and subsequent image recognition, thereby automatically capturing multiple frames of vehicle images of container vehicles entering and exiting the entrance and exit of the area to be detected and automatically performing subsequent image recognition without human intervention.

[0061] Preferably, since container vehicles are typically quite long, they require a certain period of time to pass through the entrance or exit of the inspection area. During this period, the detection device 100 can continuously detect the presence of container vehicles, and the image capture device 200 can then continuously capture multiple frames of vehicle images. It will be appreciated that not every frame of vehicle image during this period will contain the container number of the container loaded by the container vehicle. Therefore, to facilitate subsequent effective image recognition by the server 300, the image capture device 200 will send all vehicle image frames within this period of time to the server 300 for image recognition after completing the capture.

[0062] After receiving multiple frames of vehicle images sent by the image acquisition device 200, the server 300 needs to detect and identify each frame of the vehicle image. The server 300 is preferably equipped with a text contour detection algorithm and a text recognition algorithm.

[0063] The text contour detection algorithm preferably employs the FCENet algorithm, which employs a resampling mechanism to obtain a fixed number of dense points for each text contour. To ensure the uniqueness of the Fourier eigenvector, the sampling starts at the rightmost intersection of a horizontal line passing through the center of the text in the container number's text contour domain. The sampling direction is fixed to a clockwise direction, with equally spaced samples taken along the container number's text contour. Secondly, the sequence of contour sampling points in the spatial domain is embedded into the Fourier domain via a Fourier transform. Preferably, the FCENet algorithm consists of three components: a feature extractor, a feature pyramid network, and a prediction layer. The feature extractor is ResNet101, a residual network model equipped with deformable convolutions, as the feature extraction layer; the feature pyramid network serves as the neck layer (i.e., the network layer); and Fourier Contour Embedding (FCE) serves as the head layer (i.e., the prediction layer). The head layer consists of two independent branches: a classification branch and a regression branch. The classification branch predicts the text region mask and the text center region mask, while the regression branch predicts the Fourier feature vector of the text in the Fourier domain. This feature vector is then input into an inverse Fourier transform to reconstruct the text contour point sequence. This algorithm achieves highly accurate and robust contour detection of text on containers. Furthermore, the FCENet algorithm demonstrates improved efficiency and excellent generalization for text contour detection on containers of arbitrary shapes.

[0064] After detecting the container number text outline for each vehicle image frame using the aforementioned text outline detection algorithm, the aforementioned text recognition algorithm is then used to identify the text symbols within the container number text outline, thereby obtaining a container number recognition result consisting of characters and numbers. The aforementioned text recognition algorithm is preferably the RobustScanner algorithm. Due to the irregularity and diversity of the text shape and layout on the container side, as well as the fact that the layout can be curved, oriented, or twisted, there is a misalignment between the output character sequence and the two-dimensional input, which presents certain difficulties for text recognition. However, the RobustScanner algorithm can address this issue to a certain extent. It is a scene text recognition algorithm based on an attention-based encoding and decoding framework. During the decoding process of the text recognition decoder, typical character-level sequence decoders not only utilize the upper and lower symbol information of the container number, but also the position information of the container number. To suppress the side effects of attention drift, the RobustScanner algorithm adopts a new position enhancement branch and dynamically fuses its output with the output of the decoder's attention module for text recognition in container entry and exit scenarios. Specifically, it contains a position-aware module that enables the encoder to output a feature vector encoding its own spatial position, and an attention module that uses only positional cues (i.e., the current decoding time step), where dynamic fusion is achieved through a unit-level gate mechanism. In theory, the RobustScanner algorithm can decode individual characters based on the dynamic ratio between the context and positional cues of the container ID, using more positional characters when there is less contextual information, thus having strong robustness, generalization, and practicality.

[0065] Furthermore, since the server 300 simultaneously detects and identifies multiple frames of vehicle images within a certain time period, it will output multiple container number recognition results. Therefore, after detecting and identifying each frame of vehicle image, it is necessary to process the corresponding container number recognition results to obtain the container number of the container vehicle.

[0066] In a preferred embodiment of the present invention, the detection device 100 includes a controller 4 and an ultrasonic radar 5 and a laser sensor 6 connected to the controller 4;

[0067] The controller 4 is configured to output a shooting signal indicating that a container vehicle has been detected entering or exiting when the detection signals from the ultrasonic radar 5 and the laser sensor 6 are received simultaneously.

[0068] Specifically, because the ultrasonic radar 5 is based on the Doppler effect and utilizes a state-of-the-art planar antenna, it effectively suppresses interference from higher harmonics and other clutter, resulting in a wide detection range. The laser sensor 6, on the other hand, utilizes laser technology, enabling contactless, long-distance measurement. Compared to radar sensors, it is faster, more accurate, and more resistant to photoelectric interference. Therefore, in this embodiment, by simultaneously providing the ultrasonic radar 5 and the laser sensor 6, they can mutually verify each other during detection, avoiding false triggering caused by interference where only one of the ultrasonic radar 5 or the laser sensor 6 detects an external object. The two complement each other's strengths, ensuring the accuracy and breadth of triggering during container vehicle entry and exit. The two sensor types differ in detection parameters such as detection range, resolution, and angular resolution, resulting in distinct advantages and disadvantages corresponding to container vehicle detection capabilities, modeling capabilities, and weather resistance, effectively completing the triggering task for container entry and exit.

[0069] In a preferred embodiment of the present invention, the image acquisition device 200 includes multiple surveillance cameras 7 at different angles installed at the entrance and exit of the area to be detected; the result processing module 3 includes:

[0070] The first verification submodule 31 is used to perform container number verification on each container number recognition result, so as to add the container number recognition results that meet the preset container number verification rules into a first result set;

[0071] The second check submodule 32 is connected to the first check submodule 31 and is used to match the box owner code included in each box number recognition result in the first result set with at least one pre-configured standard code, and add the box number recognition result corresponding to the box owner code matching any standard code to a second result set;

[0072] The set division submodule 33 is connected to the second verification submodule 32 and is used to divide the second result set into a plurality of subsets according to the surveillance camera corresponding to each box number recognition result in the second result set;

[0073] The processing submodule 34 is connected to the set division submodule 33, and is used to obtain the corresponding single camera recognition result associated with the surveillance camera according to the recognition result of each container number in each subset, and when it is determined that there are at least two surveillance cameras associated with the same single camera recognition result, the same single camera recognition result is used as the container number of the container vehicle.

[0074] Specifically, in this embodiment, preferably, four surveillance cameras 7 with different angles are installed around the entrance and exit of the area to be detected to monitor the entry and exit of container vehicles in real time. Each surveillance camera 7 collects the video stream at the entrance and exit of the area to be detected in real time. When the detection device 100 detects that a container vehicle has entered or exited, it sends a shooting signal to each surveillance camera, and each surveillance camera collects images of the current video stream in frames as the minimum unit to obtain multiple frames of vehicle images. The image acquisition device 200 realizes high-speed image acquisition and storage of the surveillance camera 7, and preferably realizes the transmission of vehicle images through serial communication between the network video recorder (NVR) and the server 300, and then realizes target detection and recognition on the server 300.

[0075] As previously mentioned, because server 300 simultaneously detects and recognizes multiple vehicle image frames within a certain time period, it will output multiple container number recognition results. However, not every container number recognition result will contain a correct and complete container number. Therefore, after detecting and recognizing each vehicle image frame, it is necessary to use multiple intelligent verification methods to verify each container number recognition result.

[0076] The multiple intelligent verification means include container number verification and container owner code verification, wherein the container number verification is to confirm whether the corresponding container number identification result complies with the preset container number verification rules. If it does not comply, it is an invalid container number identification result. Specifically, the above-mentioned preset container number verification rules are formulated based on the composition of the standard container number. The container number adopts the international standard ISO6346 (1995) standard definition and consists of 7 digits (container registration code) and 4 English letters (container owner code). The last digit of the 7 digits is obtained by calculating the first four letters and 6 digits through the verification rules. The verification rules are conventional technical means in this field and will not be repeated here. It can be understood that the above-mentioned preset container number verification rules may include first judging whether the first four digits are English letters and whether the last seven digits are numbers, and then judging whether the last digit complies with the verification rules based on the above-mentioned verification rules.

[0077] In addition to container number verification, owner code verification is also available to further eliminate incorrect container number identification results. The owner code is a four-character uppercase code representing the container owner. This verification compares the four-digit owner code identified with the owner code in the international BIC code registry, further eliminating instances where the container number verification fails but is in fact incorrect. As needed, one or more owner codes in the international BIC code registry can be selected as standard codes for matching.

[0078] In order to reduce the missed detection rate of container vehicles in extreme cases, in this embodiment, the container number recognition result that passes the above-mentioned container number verification and box owner code verification is further expanded. In a preferred embodiment of the present invention, each frame of vehicle image is associated with the image acquisition time;

[0079] The result processing module 3 further includes a result expansion submodule 35, which is connected to the set partitioning submodule 33 and the first check submodule 31 respectively. The result expansion submodule 35 includes:

[0080] The sorting unit 351 is used to sort the box number recognition results corresponding to each frame of vehicle image in each subset according to the order of image acquisition time before the processing submodule processes the box number recognition results in each subset;

[0081] The expansion unit 352 is connected to the sorting unit 351 and is configured to, for the sorted subset, have the server sequentially count the number of occurrences of each character in at least two consecutive box number recognition results according to the sorting, then use the character with the largest number of occurrences as the character of the corresponding position to form a new recognition result, and save the new recognition result as the box number recognition result to expand the box number recognition result;

[0082] The first syndrome module 31 and the second syndrome module 32 perform container number verification and matching on the expanded container number recognition results again in sequence to update the second result set;

[0083] The set division submodule 33 divides the updated second result set into subsets again.

[0084] Specifically, in this embodiment, each subset corresponds to a surveillance camera 7, and each container number recognition result in this subset is obtained by detecting and identifying the vehicle image captured by the corresponding surveillance camera 7 and performing the above-mentioned two-step verification. When expanding the results, each subset is expanded separately. Specifically, the most frequently appearing characters in all container number recognition results in each subset are extracted and pieced together to form a new container number. Considering that simply permuting and combining the container number recognition results of each frame would greatly waste computing resources, in this embodiment, expansion is based on at least two consecutive container number recognition results, preferably based on the container number recognition results of two adjacent frames with similar time. Based on this, it is first necessary to sort the container number recognition results in each subset based on the image acquisition time. It is understandable that each frame of vehicle image is associated with a corresponding surveillance camera 7 and image acquisition time, and sorting can be based on this. Alternatively, if the image acquisition time is missing, the image acquisition time can be represented based on the distance of the image coordinates of the container number text outline in each frame. Close image coordinates indicate similar image acquisition times.

[0085] For example, if the case owner codes of two adjacent case number recognition results in one subset are ABCD and ABEF, the most common first-digit character is A, the most common second-digit character is B, the most common third-digit characters are C and E, and the most common fourth-digit characters are D and F. Therefore, the new recognition results include ABCF and ABED. It is understandable that this example only shows the expansion method of the case owner code for ease of description. The expansion method of the complete case number recognition result is similar.

[0086] Preferably, for the entire subset, taking the box number recognition results of 100 frames of vehicle images in the subset as an example, the box number recognition results in each subset can be grouped first, such as 10 as a group, and then for each two adjacent box number recognition results in each group, the expansion method of the box master code is referred to. Among them, the grouping method can be such as 1-10 as the first group, 11-20 as the second group, and so on, or such as 1-10 as the first group, 11-21 as the second group, and so on. It can be understood that when specifically grouping, the number of box number recognition results contained in each group is not limited to 10, and can be configured according to needs.

[0087] After the container number recognition result is expanded, the container number needs to be verified and matched again for the new recognition result to eliminate the erroneous new recognition result.

[0088] It is understandable that the expansion of the above-mentioned box number recognition result can also be performed before the container number verification, that is, the box number recognition result output by the text recognition module 2 is directly used for expansion. However, since the box number recognition result output by the text recognition module 2 contains erroneous recognition results and the data volume is large, further expansion will cause a huge waste of computing resources and consume a long computing time, reducing the real-time performance of this system. Therefore, in this embodiment, expansion is performed after two-step verification, and two-step verification is performed again after expansion, which greatly improves computing efficiency.

[0089] In a preferred embodiment of the present invention, each box number recognition result is associated with a corresponding confidence level;

[0090] The processing submodule 34 includes:

[0091] The first statistical unit 341 is used to count all result values ​​of the box number recognition results in each subset as the single camera recognition results associated with the corresponding surveillance camera;

[0092] a determination unit 342 connected to the first statistical unit 341 and configured to output, for each single-camera recognition result, the single-camera recognition result when it is determined that the single-camera recognition result exists in at least two subsets at the same time, and to generate a first signal when it is determined that the single-camera recognition result does not exist in at least two subsets at the same time;

[0093] The second counting unit 343 is connected to the judgment unit 342 and is configured to count, in each subset according to the first signal, a first number of box number recognition results having the same result value as the box number recognition result associated with the highest confidence level, and output the single camera recognition result associated with the subset having the highest confidence level, the highest confidence level being greater than a preset threshold, and the largest first number;

[0094] The output unit 344 is connected to the judgment unit 342 and the second statistical unit 343 respectively, and is used to use the output single camera recognition result as the container number of the container vehicle.

[0095] Specifically, in this embodiment, to further enhance the accuracy of container number detection and recognition results, and because the image acquisition device 200 includes multiple surveillance cameras 7 with different angles, the aforementioned multiple intelligent verification methods further include multi-camera verification. Multi-camera verification compares and filters the container number recognition results of vehicle images captured by each surveillance camera 7: only when the container number recognition results of vehicle images captured by at least two surveillance cameras 7 are identical will the container number recognition result be retained. Multi-camera verification further ensures the accuracy of container number detection and recognition results, eliminates the randomness and uniqueness of single-camera detection and recognition, and improves the robustness and accuracy of container number detection and recognition.

[0096] Before determining whether there are at least two surveillance cameras 7 that have the same box number recognition results for vehicle images captured by them, it is necessary to first determine the box number recognition results associated with each surveillance camera 7. Since each surveillance camera 7 captures multiple frames of vehicle images, although some erroneous box number recognition results are eliminated through two-step verification, there may still be multiple box number recognition results in the subset corresponding to each surveillance camera 7. Moreover, since the box number recognition result contains a total of 11 characters, the result values ​​of each box number recognition result (i.e., 11 characters) are not necessarily consistent, and there may be deviations in individual characters, which may result in each subset containing more than two box number recognition results. However, in order to avoid missed detection, in this embodiment, all result values ​​of each box number recognition result in the subset corresponding to each surveillance camera 7 are first retained as the single camera recognition result of each surveillance camera 7.

[0097] To be more specific, taking the example of a subset containing 10 box number recognition results, if the first result value has 6 box number recognition results and the second result value has 4 box number recognition results, then both the first result value and the second result value will be used as the box number recognition results of the surveillance camera 7, that is, the single camera recognition result.

[0098] Furthermore, if the single-camera recognition results of at least two surveillance cameras 7 are the same, the single-camera recognition result is retained. In other words, if one of the surveillance cameras 7 contains the first result value and the second result value, and the other surveillance camera 7 contains the first result value and the third result value, the first result value is retained, and so on. However, in extreme cases, there may not be at least two single-camera recognition results that are the same, that is, four surveillance cameras 7 correspond to four different container number recognition results: for example, due to the parking of other vehicles at the entrance and exit of the area to be detected, the container vehicle's entry and exit routes are irregular, and there are objects blocking the container vehicle when entering and exiting. The container number on the side and top of the container vehicle will bypass the surveillance camera 7 for collection, resulting in only a single surveillance camera 7 being able to collect the vehicle image with the container number and detect and identify it. In this case, in this embodiment, the method of single-camera confidence compensation is relied upon to solve such technical problems. Since the container number recognition result is implemented based on the neural network algorithm, each container number recognition result output by the neural network algorithm is associated with a confidence level. The single-camera confidence compensation method uses the confidence level returned by the container number recognition results to select the single-camera results that meet the following three conditions: the highest confidence level, exceeding a very high threshold (such as 0.99), and the largest number of results as supplementary results. This method addresses the problem of missed detections in the multi-camera verification method in situations such as irregular container vehicle routes and obstructed roads. The above-mentioned preset threshold is the very high threshold.

[0099] Specifically, for each surveillance camera 7, the highest confidence level among the container number recognition results in its corresponding subset is determined. It is then determined whether the confidence level is greater than a preset threshold (e.g., 0.99). If not, the single-camera recognition results of the surveillance camera 7 corresponding to the subset are discarded. If the confidence level is greater than the preset threshold, a first number of container number recognition results with the same result value as the container number recognition result corresponding to the confidence level is further counted. The single-camera recognition results of the surveillance camera 7 corresponding to the subset with the largest first number, i.e., the single-camera results, are then selected as the final container number.

[0100] In actual applications, it is possible that two container trucks (or even more trucks) enter at the same time with their head and tail connected. That is, the detection device 100 is triggered once but multiple trucks enter. Therefore, the final output result will be more than one. Based on this, in a preferred embodiment of the present invention, the processing submodule 34 further includes a time period verification unit 345, which is respectively connected to the judgment unit 342, the second statistical unit 343 and the output unit 344. The time period verification unit 345 includes:

[0101] a first check subunit 3451, configured to add all output single-camera recognition results to a third result set, and output a second signal when it is determined that the number of elements in the third result set is greater than one, and output a third signal when the number of elements is not greater than one;

[0102] The second check subunit 3452 is connected to the first check subunit 3451 and is configured to perform bit-wise comparison on each pair of the single-camera recognition results in the third result set according to the second signal, and output a fourth signal when it is determined that the character difference between the two compared single-camera recognition results is no more than one bit, and output a third signal when the character difference is more than one bit;

[0103] a third check subunit 3453 connected to the second check subunit 3452 and configured to calculate, based on the fourth signal, the respective proportions of the two compared single-camera recognition results in all single-camera recognition results, and, if it is determined that the two corresponding proportions are different, delete the single-camera recognition result with the lower proportion from the third result set; and, if the two corresponding proportions are the same, delete the associated single-camera recognition result with the lower confidence level from the third result set, thereby updating the third result set and subsequently outputting a third signal;

[0104] The fourth check subunit 3454 is connected to the first check subunit 3451, the second check subunit 3452, and the third check subunit 3453, respectively, and is configured to output the recognition results of each single camera in the third result set when it is determined based on the third signal that the comparison of the recognition results of each single camera is complete;

[0105] The output unit 344 uses the recognition results of each single camera in the third result set as the container number of the container vehicle.

[0106] Specifically, through multi-camera verification, if the single-camera recognition results of at least two surveillance cameras 7 are the same, then the single-camera recognition results are retained. In the case of four surveillance cameras, if the single-camera recognition results of two of the surveillance cameras are the same, and the single-camera recognition results of the other two surveillance cameras are the same, two single-camera recognition results will also be retained, that is, it is necessary to confirm whether the multiple retained single-camera recognition results correspond to one vehicle or two vehicles. In this embodiment, a simultaneous screening algorithm is designed to address this situation. The simultaneous screening algorithm compares the container numbers of the same time period with each other, distinguishes container numbers that are particularly similar (only one character different), and then uses the number as the judgment basis to retain the result with the larger number and exclude other particularly similar recognition results; if there are two groups with the same number, the recognition confidence level is used as the judgment basis, and only the result with the highest recognition confidence level is output as the only output. The simultaneous screening algorithm can also address the situation where character recognition errors are caused by special camera angle sampling, which greatly improves and solves this technical problem.

[0107] Taking the container owner code in the container number recognition results as an example, for example, if ABCD and ABEF differ by two characters, they are considered to be two vehicles, and both results are retained. For example, if ABCD and ABCF differ by only one character, and there are only two recognition results from a single camera, the proportions of the two are the same. In this case, the one with the higher confidence level is considered the correct container number. In other words, the final output may be one container number or more than one container number.

[0108] In a preferred embodiment of the present invention, the server 300 further includes a data optimization module 8, connected to the contour recognition module 1 and the text recognition module 2, respectively, for counting a second number of container number text contours in each frame of the vehicle image, and discarding at least the container number text contour in the corresponding vehicle image when the second number is less than a preset value, and retaining the container number text contour in the corresponding vehicle image when the second number is not less than the preset value.

[0109] The text recognition module 2 recognizes the text symbols in the container number text outline corresponding to each frame of the retained vehicle image.

[0110] Specifically, in this embodiment, the preset value is preferably 3. When the number of detected container number text contours is less than three, frame skipping is performed, i.e., at least the current frame of the vehicle image is skipped, and subsequent processes such as text symbol recognition are skipped. Preferably, the detection of the container number text contours for multiple consecutive frames of vehicle images may also be skipped, such as skipping two consecutive frames. The specific number of skipped frames can be configured as needed. Only when three or more container number text contours are detected does the normal process continue, further improving process efficiency, reducing time and space complexity, and accelerating the speed of container number detection and recognition.

[0111] Furthermore, while considering the running speed of container number detection and recognition, the accuracy of container number recognition also needs to be considered. Based on this, backtracking processing is required on the basis of the above-mentioned frame skipping processing. For example, if the current frame is the first frame, and the number of container number text contours corresponding to the vehicle image of the first frame is less than three, the subsequent two consecutive frames are skipped as an example, that is, the vehicle images of the second and third frames are not processed at all, and the container text contour is directly recognized on the vehicle image of the fourth frame. If the number of container text contours corresponding to the vehicle image of the fourth frame is also less than three, the vehicle images of the fifth and sixth frames are skipped, and the container text contour is directly recognized on the vehicle image of the seventh frame, and so on. If the number of container text contours of the vehicle image of the fourth frame is not less than three, backtracking processing is required, that is, the container text contour needs to be recognized on the skipped second and third frames. If the number of container text contours of the second or third frame is not less than three, a series of subsequent processes such as text symbol recognition are performed on the corresponding frame and the fourth frame, and then the container text contour is recognized on the vehicle image of the fifth frame, and so on.

[0112] In a preferred embodiment of the present invention, the server 300 further includes a state detection module 9, connected to the contour recognition module 1, for respectively counting the text contour detection coordinates of the container number text contour in each frame of the vehicle image in the image coordinate system, and processing the corresponding text contour detection coordinates corresponding to multiple consecutive frames of vehicle images to obtain the corresponding driving direction of the container vehicle to characterize the entry and exit status of the container vehicle.

[0113] Specifically, the entry and exit trajectories of container vehicles are crucial for warehouse management. Conventional methods for determining vehicle entry and exit behavior include using geomagnetic and infrared sensors, image recognition based on license plate detection, and wireless sensing using ID cards. Some of these methods are difficult to install, can only be used in specific scenarios, and are costly. In this embodiment, during the container number recognition process, vehicle entry and exit status can be detected simultaneously based on the changing trends of text contour detection coordinates, eliminating the need for additional equipment installation and significantly reducing costs.

[0114] In a preferred embodiment of the present invention, a visualization platform 400 is further included, connected to the server 300, for visually displaying at least one of each frame of vehicle image, the container number of the corresponding container vehicle, and the entry and exit status of the container vehicle.

[0115] Specifically, container data visualization is primarily based on modern information networks and utilizes cloud computing technology as its innovation. In this embodiment, a container entry and exit data visualization platform 400 is provided, providing a more convenient way for company supervisors and administrators to manage container storage and collect data, thereby minimizing resource waste and improving resource utilization. As a preferred implementation, the visualization platform can be developed using the SpringBoot+SpringMvc+Mybatis framework, leveraging cloud computing technology to achieve capacity enhancement and technological innovation.

[0116] Further preferably, the visualization platform 400 can provide an account login port to enable the company's loading supervisor to log in to the loading supervisor account, and can also provide a container number query port to meet the needs of container ID query. It can also provide functional modules such as container entry and exit time management and container entry and exit image query; providing a more efficient and convenient online management mode for loading supervisors and company administrators.

[0117] The present invention also provides a method for intelligent detection and identification of container numbers, which is applied to the above-mentioned intelligent detection and identification system, such as Figure 2 As shown, the intelligent detection and identification method includes:

[0118] Step S1: The detection device continuously outputs a shooting signal when detecting a container vehicle entering or exiting the entrance or exit of the detection area;

[0119] Step S2: The image acquisition device continuously acquires images of the corresponding container vehicle according to the shooting signal to obtain multiple frames of vehicle images and sends them to the server;

[0120] Step S3: The server performs container number contour recognition on each frame of the vehicle image to obtain the container number text contour in each frame of the vehicle image;

[0121] Step S4: The server recognizes the text symbols in the container number text outline corresponding to each frame of the vehicle image to obtain the container number recognition result in each frame of the vehicle image;

[0122] Step S5: The server obtains the container number of the container vehicle according to the identification results of each container number.

[0123] The above description is only a preferred embodiment of the present invention and does not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the content of this specification and illustrations should be included in the protection scope of the present invention.

Claims

1. An intelligent detection and identification system for container numbers, characterized in that: include: A detection device is provided at an entrance or exit of an area to be detected, and is used to continuously output a shooting signal when a container vehicle is detected entering or exiting the entrance or exit of the area to be detected; An image acquisition device is provided at the entrance or exit of the area to be detected, and is used to continuously acquire images of the corresponding container vehicle according to the shooting signal to obtain multiple frames of vehicle images and output them; A server is connected to the image acquisition device, and the server includes: A contour recognition module is used to perform container number contour recognition on each frame of the vehicle image to obtain the container number text contour in each frame of the vehicle image; a text recognition module, connected to the contour recognition module, for respectively recognizing the text symbols in the container number text contour corresponding to each frame of the vehicle image, and obtaining a container number recognition result in each frame of the vehicle image; A result processing module, connected to the text recognition module, for obtaining the container number of the container vehicle according to each of the container number recognition results; Wherein, the image acquisition device includes multiple surveillance cameras at different angles installed at the entrance and exit of the area to be detected; and the result processing module includes: a first verification submodule, configured to perform container number verification on each of the container number recognition results, so as to add the container number recognition results that meet the preset container number verification rules into a first result set; a second check submodule, connected to the first check submodule, configured to match the case owner code included in each case number recognition result in the first result set with at least one pre-configured standard code, and add the case number recognition result corresponding to the case owner code that matches any of the standard codes to a second result set; a set division submodule, connected to the second verification submodule, for dividing the second result set into a plurality of subsets according to the surveillance camera corresponding to each of the box number recognition results in the second result set; A processing submodule, connected to the set division submodule, is used to obtain the corresponding single camera recognition result associated with the surveillance camera according to the container number recognition results in each of the subsets, and when it is determined that there are at least two of the single camera recognition results associated with the surveillance cameras that are the same, the same single camera recognition result is used as the container number of the container vehicle.

2. The intelligent detection and identification system according to claim 1, characterized in that: The detection device includes a controller and an ultrasonic radar and a laser sensor connected to the controller; The controller is configured to output the shooting signal indicating that a container vehicle is detected to be entering or exiting when the detection signals from the ultrasonic radar and the laser sensor are received simultaneously.

3. The intelligent detection and identification system according to claim 1, characterized in that: Each frame of the vehicle image is associated with an image acquisition time; The result processing module further includes a result expansion submodule, which is connected to the set partitioning submodule and the first check submodule respectively, and the result expansion submodule includes: a sorting unit, configured to sort the box number recognition results corresponding to each frame of the vehicle image in each subset according to the order of the image acquisition time before the processing submodule processes the box number recognition results in each subset; an expansion unit connected to the sorting unit, configured to, for the sorted subset, have the server sequentially count the number of occurrences of each character in at least two consecutive box number recognition results according to the sorting, then use the character with the largest number of occurrences as the character of the corresponding position to form a new recognition result, and save the new recognition result as the box number recognition result to expand the box number recognition result; The first syndrome submodule and the second syndrome submodule again perform container number verification and matching on the expanded container number recognition result in sequence to update the second result set; The set division submodule divides the updated second result set into subsets again.

4. The intelligent detection and identification system according to claim 1, characterized in that: Each of the box number recognition results is associated with a corresponding confidence level; Then the processing submodule includes: A first statistical unit is configured to, for each subset, respectively count all result values ​​of the box number recognition results in the subset as single-camera recognition results associated with the corresponding surveillance camera; a determination unit, connected to the first statistical unit, configured to, for each of the single-camera recognition results, output the single-camera recognition result when determining that the single-camera recognition result exists in at least two of the subsets simultaneously, and generate a first signal when determining that the single-camera recognition result does not exist in at least two of the subsets simultaneously; a second statistical unit, connected to the judgment unit, for respectively counting, in each of the subsets, a first number of the box number recognition results having the same result value as the box number recognition result associated with the highest confidence level according to the first signal, and outputting the single camera recognition result associated with the subset having the highest confidence level, the highest confidence level being greater than a preset threshold, and the largest first number; An output unit is connected to the judgment unit and the second statistical unit respectively, and is used to use the output single camera recognition result as the container number of the container vehicle.

5. The intelligent detection and identification system according to claim 4, characterized in that: The processing submodule further includes a time period verification unit, which is connected to the judgment unit, the second statistical unit and the output unit respectively, and the time period verification unit includes: a first check subunit, configured to add all the output single-camera recognition results to a third result set, and output a second signal when it is determined that the number of elements in the third result set is greater than one, and output a third signal when the number of elements is not greater than one; a second check subunit, connected to the first check subunit, configured to perform bit-wise comparison on each of the single-camera recognition results in the third result set in sequence according to the second signal, and output a fourth signal when it is determined that the difference in characters between the two compared single-camera recognition results is no greater than one bit, and output the third signal when the difference in characters is greater than one bit; a third check subunit, connected to the second check subunit, configured to calculate, based on the fourth signal, the respective proportions of the two compared single-camera recognition results in all the single-camera recognition results, and, if it is determined that the two corresponding proportions are different, delete the single-camera recognition result with the lower proportion from the third result set; and, if the two proportions are the same, delete the associated single-camera recognition result with the lower confidence level from the third result set, so as to update the third result set and subsequently output the third signal; a fourth check subunit, connected to the first check subunit, the second check subunit, and the third check subunit, respectively, and configured to output each of the single-camera recognition results in the third result set when it is determined based on the third signal that the comparison of each of the single-camera recognition results is complete; The output unit uses each of the single-camera recognition results in the third result set as the container number of the container vehicle.

6. The intelligent detection and identification system according to claim 1, characterized in that: The server further includes a data optimization module, connected to the contour recognition module and the text recognition module, respectively, for counting a second number of the container number text contours in each frame of the vehicle image, and discarding at least the container number text contour in the corresponding vehicle image when it is determined that the second number is less than a preset value, and retaining the container number text contour in the corresponding vehicle image when the second number is not less than the preset value; The text recognition module recognizes the text symbols in the container number text outline corresponding to each frame of the retained vehicle image.

7. The intelligent detection and identification system according to claim 1, characterized in that: The server also includes a state detection module connected to the contour recognition module, which is used to count the text contour detection coordinates of the container number text contour in each frame of the vehicle image in the image coordinate system, and obtain the corresponding driving direction of the container vehicle based on the text contour detection coordinates corresponding to multiple consecutive frames of the vehicle image to represent the entry and exit status of the container vehicle.

8. The intelligent detection and identification system according to claim 7, characterized in that: It also includes a visualization platform connected to the server, which is used to visualize at least one of the vehicle image of each frame, the container number of the corresponding container vehicle, and the entry and exit status of the container vehicle.

9. A method for intelligent detection and identification of container numbers, characterized in that: Applied to the intelligent detection and identification system according to any one of claims 1 to 8, the intelligent detection and identification method includes: Step S1, the detection device continuously outputs a shooting signal when detecting a container vehicle entering or exiting the entrance or exit of the area to be detected; Step S2, the image acquisition device continuously acquires the corresponding image of the container vehicle according to the shooting signal to obtain multiple frames of vehicle images and sends them to the server; Step S3, the server performs container number contour recognition on each frame of the vehicle image to obtain the container number text contour in each frame of the vehicle image; Step S4, the server recognizes the text symbols in the container number text outline corresponding to each frame of the vehicle image, and obtains the container number recognition result in each frame of the vehicle image; In step S5, the server obtains the container number of the container vehicle according to the container number recognition results.

Images