Goods theft risk identification method and system
By analyzing image data of the cargo compartment area in the cloud and on the vehicle, calculating scores for camera obstruction intent and theft intent, and filtering out suspicious persons, the problem of inaccurate theft intent assessment in existing technologies is solved, and accurate theft risk identification is achieved.
Patent Information
- Application Number
- CN202511155968.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-18
- Publication Date
- 2025-12-12
Smart Images
Figure CN121119682A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of freight security monitoring technology, and in particular to a method and system for identifying the risk of cargo theft. Background Technology
[0002] With the rapid development of modern logistics, the security of goods during transportation has become increasingly prominent. Long-haul freight vehicles, due to their long routes, complex environments, and unattended stops, are frequently at risk of cargo theft. To ensure cargo safety and reduce economic losses, installing video surveillance systems on freight vehicles has become a common technological approach. However, existing vehicle-mounted video surveillance systems cannot accurately distinguish between malicious and routine camera obstruction, thus failing to accurately identify theft intent and resulting in low accuracy in assessing theft intent.
[0003] Therefore, improving the accuracy of assessing theft intent has become an urgent technical problem to be solved. Summary of the Invention
[0004] This invention provides a method, system, electronic device, storage medium, and computer program product for identifying the risk of theft of goods, in order to overcome the deficiencies in the prior art and improve the accuracy of assessing theft intent.
[0005] This invention provides a method for identifying theft risks of goods, applied in the cloud, including: Receive status data sent by the vehicle-mounted terminal, wherein the status data is obtained by analyzing image data of the cargo compartment area collected by the vehicle-mounted terminal; Extract multiple first features corresponding to each first person in the cargo compartment area from the state data, and calculate the camera occlusion intention score of each first person based on the multiple first features; Based on the scores for each camera obstruction intent, a second person is determined from the first person; Based on the status data, a third person who interacted with the second person is identified; Extract multiple second features corresponding to each third person in the cargo compartment area from the status data, and calculate the theft intent score of each third person based on the multiple second features; Based on the scores of the stated theft intent, the theft risk identification results are determined.
[0006] According to a cargo theft risk identification method provided by the present invention, the step of extracting multiple first features corresponding to each first person in the cargo compartment area from the status data, and calculating the camera occlusion intention score of each first person based on the multiple first features, includes: For any first person, the speed at which the first person approaches the camera is extracted from the state data to obtain a speed feature; The first temporal feature is obtained by extracting the duration during which the first person continuously obstructs the camera from the state data. The distance between the first person and the camera is extracted from the state data to obtain distance features; First action data of the first person is extracted from the state data to obtain a first action feature; wherein, when the first action data matches a preset action, the value of the first action feature is 1; when the first action data does not match the preset action, the value of the first action feature is 0. Extract time period features from the status data; wherein, when it is a loading / unloading period, the value of the time period feature is 1; when it is not a loading / unloading period, the value of the time period feature is 0; The camera occlusion intention score is calculated based on the speed feature, the first time feature, the distance feature, the first action feature, and the time period feature.
[0007] According to a cargo theft risk identification method provided by the present invention, the step of extracting multiple second features corresponding to each third person in the cargo compartment area from the status data, and calculating a theft intent score for each third person based on the multiple second features, includes: For any third person, the frequency of the third person's contact with the preset item is extracted from the state data to obtain frequency features; The probability of the third person carrying a preset theft tool is extracted from the state data to obtain a probability feature; The duration of the third person's stay in the cargo compartment warning area is extracted from the status data, and the duration of stay is standardized to obtain a second time feature; Extract the second action of the third person from the state data to obtain the second action feature; Calculate the similarity between the second action feature and the historical theft action features to obtain the theft correlation feature; The theft intent score is calculated based on the frequency feature, the probability feature, the second time feature, and the theft correlation feature.
[0008] The method for identifying theft risks of goods provided by the present invention further includes: The theft risk identification results are sent to the vehicle-mounted terminal.
[0009] This invention provides a method for identifying theft risks of goods, applied to a vehicle-mounted terminal, comprising: Collect image data of the cargo compartment area; The image data is analyzed to obtain status data; The status data is sent to the cloud so that the cloud can determine the theft risk identification result; the theft risk identification result is obtained by extracting multiple first features corresponding to each first person in the cargo compartment area from the status data, and calculating the camera obstruction intention score of each first person based on the multiple first features; determining a second person from the first persons based on the camera obstruction intention score; determining a third person who interacts with the second person based on the status data; extracting multiple second features corresponding to each third person in the cargo compartment area from the status data, and calculating the theft intention score of each third person based on the multiple second features; and determining the third person based on the theft intention score.
[0010] According to a method for identifying the risk of cargo theft provided by the present invention, the step of analyzing the image data to obtain status data includes: Perform personnel detection on the image data to obtain the intersection-union ratio (IoU) of the personnel detection bounding box and the cargo compartment warning area coordinates; Based on the image data, determine the degree of occlusion and the deflection angle of the camera; Determine whether any one of the intersection-union ratio, the degree of occlusion, and the deflection angle meets a preset condition, and when the preset condition is met, encapsulate the coordinates of the personnel detection box, the degree of occlusion, the deflection angle, the acquisition time, the geographical location, and the image data into the state data.
[0011] The method for identifying theft risks of goods provided by the present invention further includes: When the degree of occlusion is greater than a first preset threshold or the deflection angle is greater than a second preset threshold, the alarm device on the vehicle is triggered.
[0012] The method for identifying theft risks of goods provided by the present invention further includes: Receive the theft risk identification result sent by the cloud; When the theft risk identification result is high-risk, the alarm device on the vehicle is triggered.
[0013] This invention also provides a cargo theft risk identification system, applied in the cloud, comprising: The cloud receiving module is used to receive status data sent by the vehicle-mounted terminal. The status data is obtained by analyzing the image data of the cargo compartment area collected by the vehicle-mounted terminal. The cloud processing module is used to extract multiple first features corresponding to each first person in the cargo compartment area from the status data, and calculate the camera occlusion intention score of each first person based on the multiple first features. The cloud processing module is also used to determine the second person from the first person based on the occlusion intention scores of each of the cameras; The cloud processing module is also used to determine, based on the status data, a third person who has interacted with the second person; The cloud processing module is also used to extract multiple second features corresponding to each of the third persons in the cargo compartment area from the status data, and to calculate the theft intent score of each of the third persons based on the multiple second features; The cloud processing module is also used to determine the theft risk identification result based on the theft intent scores of each item.
[0014] This invention also provides a cargo theft risk identification system, applied to a vehicle-mounted terminal, comprising: The vehicle-mounted data acquisition module is used to collect image data of the cargo compartment area; The vehicle-mounted processing module is used to analyze the image data to obtain status data; The vehicle-mounted sending module is used to send the status data to the cloud so that the cloud can determine the theft risk identification result; the theft risk identification result is obtained by extracting multiple first features corresponding to each first person in the cargo compartment area from the status data, and calculating the camera obstruction intention score of each first person based on the multiple first features; determining a second person from the first persons based on the camera obstruction intention score; determining a third person who interacts with the second person based on the status data; extracting multiple second features corresponding to each third person in the cargo compartment area from the status data, and calculating the theft intention score of each third person based on the multiple second features; and determining the theft risk identification result based on the theft intention score.
[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the cargo theft risk identification method as described above.
[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the cargo theft risk identification method as described above.
[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the cargo theft risk identification method as described above.
[0018] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: By receiving initial screening status data from the vehicle-mounted terminal, the amount of data transmission was significantly reduced, achieving effective collaboration between edge and cloud resources. By analyzing personnel behavioral characteristics to calculate camera obstruction intent scores and using this to screen core suspicious individuals, the system accurately distinguished between malicious and routine camera obstruction, improving the accuracy of anomaly detection. Further identification of potential individuals interacting with core suspicious individuals avoided limitations in the analysis. Finally, by quantifying and assessing the theft-related behaviors of all potential individuals and determining the theft risk level, a precise assessment of theft risk was achieved. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0020] Figure 1 This is one of the flowcharts of the cargo theft risk identification method provided by the present invention.
[0021] Figure 2 This is the second flowchart of the cargo theft risk identification method provided by the present invention.
[0022] Figure 3 This is one of the structural schematic diagrams of the cargo theft risk identification system provided by the present invention.
[0023] Figure 4 This is the second structural schematic diagram of the cargo theft risk identification system provided by the present invention.
[0024] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0026] It should be noted that in the description of this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. The terms "upper," "lower," etc., indicating orientation or positional relationships according to the accompanying drawings, are only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0027] The terms "first," "second," etc., used in this invention are used to distinguish similar objects, not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, a first object can be one or more. Furthermore, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.
[0028] The following is combined with Figures 1-5 This invention describes the cargo theft risk identification method, system, electronic device, storage medium, and computer program product provided by this invention.
[0029] This invention provides a method for identifying theft risks of goods, applied in the cloud, with reference to... Figure 1 , Figure 1 This is one of the flowcharts illustrating the cargo theft risk identification method provided by the present invention, such as... Figure 1 As shown, steps 101 to 106 are included: Step 101: Receive status data sent by the vehicle-mounted terminal. The status data is obtained by analyzing the image data of the cargo compartment area collected by the vehicle-mounted terminal.
[0030] Specifically, status data is a structured data packet, not a raw video stream. It contains key information extracted after preliminary analysis by the vehicle-mounted device. Specifically, status data may include: the coordinates of the person detection bounding box, used to pinpoint the location of people in the image; device status parameters, such as camera occlusion rate and deflection angle, reflecting the physical state of the acquisition device; the vehicle's GPS location information and acquisition timestamp, used to record the spatiotemporal context of the event; and compressed keyframe images. The vehicle-mounted device encapsulates this information before it is received by the cloud.
[0031] One specific implementation of this step involves the cloud platform continuously monitoring the network port to prepare for receiving data. The vehicle-mounted device does not continuously send data; instead, it initiates uploading only after preset trigger conditions are met. For example, when the vehicle-mounted device detects someone entering a preset cargo compartment warning area, or detects an anomaly in the physical state of the camera (such as obstruction or angular deviation), it sends the generated status data at that moment via a low-overhead IoT communication protocol such as MQTT. The cloud platform receives the data packet carried by this protocol, thus completing the data reception process.
[0032] The specific method for collecting image data of the cargo compartment area on the vehicle-mounted terminal, analyzing the resulting status data, and sending the status data to the cloud will be described in detail in subsequent embodiments, and therefore will not be described in detail here.
[0033] Step 102: Extract multiple first features corresponding to each first person in the cargo compartment area from the status data, and calculate the camera occlusion intention score of each first person based on the multiple first features.
[0034] Since the causes of abnormal camera status (such as obstruction) are diverse, they may be caused by non-human factors such as bumps and dust, or by unintentional or malicious actions by personnel. In order to accurately assess the risk, it is necessary to precisely distinguish malicious obstruction behavior with subjective intent. Therefore, after receiving the status data, step 102 provides objective evidence for judging the obstruction intent by quantitatively analyzing the characteristics of personnel behavior.
[0035] In one possible implementation, step 102 specifically includes the following steps: For any first person, extract the speed at which the first person approaches the camera from the state data to obtain the speed feature; The duration during which the first person continuously obstructs the camera is extracted from the state data to obtain the first temporal feature; The distance between the first person and the camera is extracted from the status data to obtain distance features; The first action data of the first person is extracted from the status data to obtain the first action feature; wherein, when the first action data matches the preset action, the value of the first action feature is 1; when the first action data does not match the preset action, the value of the first action feature is 0. Extract time period features from the status data; where the value of the time period feature is 1 when it is in the loading / unloading period and 0 when it is not in the loading / unloading period. The camera occlusion intent score is calculated based on speed characteristics, first time characteristics, distance characteristics, first action characteristics, and time period characteristics.
[0036] Specifically, the "first person" refers to any individual identified within the cargo compartment area by the vehicle-mounted terminal or the cloud. Multiple "first features" are a set of specific indicators used to assess the person's intention to obstruct the camera. These features include: speed feature (the speed at which the first person approaches the camera); time feature (the duration for which the first person continuously obstructs the camera); distance feature (the shortest physical distance between the first person and the camera); action feature (a binary feature, with a value of 1 when the person is detected to have raised a hand or held an object close to the camera, otherwise 0); and time period feature (also a binary feature, used to determine whether the event occurred during a preset normal loading / unloading period, with a value of 0 if it occurred outside of this period, otherwise 1). Finally, the camera obstruction intention score is a numerical value calculated by combining the above multiple first features, used to quantify the probability that the first person actively obstructs the camera.
[0037] The detailed implementation of this process is as follows. First, for any first person, multiple first features are extracted. To obtain velocity features, consecutive keyframe images in the state data are analyzed. By comparing the center coordinates of the person's detection box in two consecutive frames and combining this with the timestamp interval between the two frames, the person's movement velocity on the image plane can be calculated. Combining this with the change in the detection box size, the movement in the depth direction can be estimated, thus obtaining a quantized value representing the velocity of approaching or moving away from the camera.
[0038] Next, to obtain the first-time features, the start and end of an occlusion event are recorded. Timing begins when the image analysis algorithm determines that the proportion of the camera's image area obscured exceeds a preset threshold; timing stops when the obstruction proportion falls below the threshold. The total duration recorded in this process is the continuous occlusion time.
[0039] Furthermore, to obtain distance features, the distance is estimated based on the size of the detection box containing the person in the image. Generally, the larger the detection box area, the closer the person is to the camera. Through pre-calibration, the pixel area of the detection box can be mapped to an approximate physical distance value.
[0040] Furthermore, to obtain the first motion feature, a deeper analysis of the keyframe images is performed. A pre-trained human pose estimation algorithm can be applied, which can identify the skeletal key points of a person in the image, such as the positions of the head, shoulders, and hands. By determining whether the hand key points are raised to a specific area (such as near the camera) or whether there is a holding action, it is determined whether the first motion data matches a preset motion. If a match is found, the first motion feature is assigned a value of 1.
[0041] Furthermore, to obtain time-period characteristics, the collection timestamps in the status data are compared with a vehicle operation plan table pre-stored in the cloud. This plan table defines the normal loading and unloading operation periods for vehicles at specific locations and times. If the collection timestamp falls within the preset loading and unloading period, the time-period characteristic is assigned a value of 1; otherwise, it is assigned a value of 0.
[0042] After extracting all the above features, the camera occlusion intent score is calculated based on the speed feature, first time feature, distance feature, first action feature, and time period feature. A specific calculation method involves applying a preset active occlusion scoring function. In this invention, the active occlusion scoring function is specifically expressed by the following formula: in, For the i-th first person, Score the camera occlusion intention of the i-th first person. Let i be the speed characteristic of the first person. For the first time feature of the i-th first person, Let D be the distance feature of the i-th first person, and D be a preset constant value. Let be the first action feature of the i-th first person, where the value of the first action feature is 1 when the first action data matches the preset action; and the value of the first action feature is 0 when the first action data does not match the preset action. The time period characteristic is defined as follows: when it is during the loading / unloading period, the value of the time period characteristic is 1; when it is not during the loading / unloading period, the value of the time period characteristic is 0. , , , , These are preset weighting factors, which sum to 1. They are obtained based on empirical data or model training and represent the importance of each feature in determining the intention to actively occlude. This means that when the speed exceeds 2 m / s, it is calculated as 2 m / s (to avoid the influence of extreme values). If the occlusion lasts for more than 10 seconds, it is counted as 1.
[0043] Finally, by substituting all the extracted feature values into this function, a quantified camera occlusion intention score can be calculated.
[0044] Step 103: Based on the scores for each camera's attempt to block, determine the second person from the first person.
[0045] After calculating the camera obstruction intent score in the aforementioned steps, step 103 is executed: based on the camera obstruction intent scores, the second person is determined from the first group of personnel. This step is performed because not all personnel appearing in the cargo compartment area pose the same risk. To concentrate analysis resources and improve the accuracy of subsequent theft intent assessment, it is necessary to screen from all detected personnel (the first group of personnel) those whose behavior is most suspicious and who are most likely to actively interfere with the monitoring equipment. These individuals are the core targets of subsequent risk analysis.
[0046] The second group of people refers to specific individuals identified through screening as having the intention to actively obstruct the camera view. This identification process is not subjective but based on a score calculated in the previous step indicating such intention. By setting a clear threshold, the score can be quantified into a concrete judgment result, thereby enabling the screening of individuals.
[0047] One specific implementation of this step is to apply a preset judgment threshold. For example, an active occlusion score threshold can be set. For example, 0.6. The camera occlusion intent score calculated by each first person is compared to this threshold. If a first person's score is greater than or equal to... If the score is below the threshold, the individual is identified as the second person, i.e., a suspicious person with an intention to actively obscure information. All first-person individuals with scores below this threshold are temporarily excluded from the core analysis. In another possible implementation, a fuzzy range can be set, for example, individuals with scores between 0.3 and 0.6 can be marked as "to be observed," and further cross-validation can be performed using other information to improve the flexibility and accuracy of the judgment.
[0048] Step 104: Based on the status data, identify the third person who has interacted with the second person.
[0049] Specifically, the third party is the set of individuals analyzed in the subsequent assessment of theft intent. This set includes not only the second party identified in the previous step, but also other individuals who interact with the second party. Interaction refers to the temporal and spatial connections between different individuals, suggesting that they may belong to the same group or be engaged in collaborative activities.
[0050] One specific implementation of this step involves making judgments based on the detection box coordinates and timestamp information of each person contained in the state data. First, all identified second persons are directly included in the initial set of third persons. Then, all other first persons in the scene, excluding the second persons, are traversed. For each other first person, the spatial distance between their detection box and the detection box of any second person is calculated. This can be achieved by calculating the Euclidean distance between the center points of the two detection boxes, or by calculating the shortest distance between the edges of the two detection boxes. When this spatial distance remains less than a preset distance threshold, such as 2 meters, for a continuous period of time, for example, more than 3 seconds, it is determined that there is interaction between the two persons. All first persons determined to have interaction are also added to the set of third persons. Finally, a set of third persons is formed, containing all core suspicious individuals and their potential accomplices.
[0051] Step 105: Extract multiple second features corresponding to each third person in the cargo compartment area from the status data, and calculate the theft intent score of each third person based on the multiple second features.
[0052] After identifying suspicious groups, further in-depth analysis of their behavior is needed to determine whether they have a genuine intent to steal. This step aims to transform the vague and subjective concept of "theft intent" into an objective and calculable score by quantifying specific behavioral characteristics, thereby providing a direct basis for the final risk level determination.
[0053] In one possible implementation, step 105 specifically includes the following steps: For any third person, extract the frequency of the third person's contact with the preset item from the state data to obtain frequency characteristics; The probability feature is obtained by extracting the probability of a third person carrying a pre-set theft tool from the state data; The duration of a third person's stay in the cargo compartment's security area is extracted from the status data, and the duration is standardized to obtain a second time feature. Extract the second action of the third person from the state data to obtain the second action feature; Calculate the similarity between the second action feature and the historical theft action features to obtain the theft correlation feature; The theft intent score is calculated based on frequency characteristics, probability characteristics, second time characteristics, and theft correlation characteristics.
[0054] Specifically, the multiple second features are a set of key indicators specifically designed to assess theft intent. These features differ from the first features used to determine concealment intent, focusing more on actions directly related to theft. These include: frequency features, indicating the frequency with which the third person touches key items in the cargo compartment (such as locks or tarpaulins); probability features, indicating the probability, analyzed by a target detection model, of whether the person is carrying pre-set theft tools such as crowbars or pliers; second time features, indicating the standardized duration of the person's stay in the guarded area, such as the cargo compartment door; and theft correlation features, obtained by calculating the similarity between the current person's actions and historical theft actions. Finally, the theft intent score is a quantitative value calculated by integrating these multiple second features, directly reflecting the likelihood of the third person committing theft.
[0055] The process begins by extracting multiple secondary features for any third person. First, the frequency of contact between the third person and preset items is extracted from the state data, yielding frequency features. Here, the preset items specifically refer to key parts of the cargo compartment, such as door locks, latches, and the edges of the tarpaulin.
[0056] Next, the probability of a third party carrying a pre-set theft tool is extracted from the state data to obtain probability features. The pre-set theft tools here include common tools used in crimes such as crowbars, bolt cutters, and scissors.
[0057] Furthermore, the duration of a third person's stay in the cargo compartment's security area is extracted from the status data, and the duration is standardized to obtain a second time feature.
[0058] Furthermore, the second action of the third person is extracted from the state data to obtain the second action feature. Here, the second action can be a sequence of hand movement trajectories or a temporal change in the person's body posture.
[0059] Furthermore, the similarity between the second action feature and the historical theft action features is calculated to obtain the theft correlation feature. Here, the historical theft action features are standard action templates stored in the cloud feature library, such as "prying", "climbing", and "hiding".
[0060] After extracting the above features, the theft intent score is calculated based on frequency features, probability features, second temporal features, and theft correlation features. One detailed implementation involves first quantizing features using image analysis techniques. For example, the contact state is continuously recorded by calculating the intersection-union ratio (IoU) between the detection bounding box of the person's hand and the detection bounding box of a preset item area, and the proportion of total contact time to total observation time is calculated as the frequency feature. Then, a pre-trained object detection model is called to analyze the person's hand or its surrounding area in the image; the confidence level of categories such as "crowbar" output by the model is the probability feature. Next, the accumulated dwell time is divided by a saturation value (e.g., 120 seconds) for normalization to obtain the second temporal feature. Further, the coordinate sequence of key points on the person's hand is extracted as the second action feature, and then the Dynamic Time Warping (DTW) algorithm is used to calculate the morphological similarity between this coordinate sequence and standard template sequences such as "crowding" in the feature library; the result is the theft correlation feature. Finally, the quantified feature values are substituted into a theft intent scoring function, which is specifically expressed by the following formula in this invention: in, For the j-th third person, Score the theft intent of the j-th third person. Let be the frequency characteristic of the j-th third person. The value of the frequency characteristic ranges from 0 to 1, where 1 indicates continuous contact. Let be the probability characteristic of the j-th third person. The value of the probability characteristic is between 0 and 1, where 1 indicates that the person is clearly carrying a pre-set theft tool. Let be the second time feature of the j-th third person. The value of the second time feature ranges from 0 to 1. , The dwell time is 120 seconds, which is the saturation value. Let be the theft correlation feature of the j-th third party, and the value of the theft correlation feature is between 0 and 1; , , , These are preset weighting factors, which sum to 1. They are obtained based on empirical data or model training and represent the importance of each feature in judging the intent to steal.
[0061] Step 106: Determine the theft risk identification results based on the scores of each theft intent.
[0062] Specifically, the theft risk identification result is the qualitative judgment that is the final output of this invention. This result is usually expressed as a specific risk level, such as "high risk," "medium risk," or "low risk." This result integrates information from all the preceding analysis steps and is the final assessment of the current security status of the cargo compartment area.
[0063] One specific implementation of this step involves applying a set of preset risk level thresholds. First, the theft intent score calculated for each third party is compared to this set of thresholds. For example, a theft intent score greater than or equal to 0.7 can be considered high risk; a score between 0.4 (inclusive) and 0.7 (exclusive) can be considered medium risk; and a score less than 0.4 can be considered low risk. During the judgment, the scores of all third parties are iterated. If any one of them reaches the "high risk" threshold, the overall theft risk identification result for this event can be determined as "high risk." If no one reaches high risk, but someone reaches medium risk, the result is determined as "medium risk." Only when all scores are below the low risk threshold is the result "low risk." Finally, this unique risk identification result, representing the highest threat level in the entire scenario, is output.
[0064] In one possible implementation, after step 106, the method further includes: The theft risk identification results are sent to the vehicle terminal.
[0065] Specifically, after generating the theft risk identification result, the cloud immediately encapsulates it into a data packet. To ensure communication security and prevent information from being tampered with or stolen, the data packet is encrypted before transmission. Subsequently, the encrypted data packet is sent to the corresponding vehicle terminal via the same communication protocol (such as MQTT) used when uploading data from the vehicle terminal, or through a dedicated downlink command channel. Upon receiving this data packet, the vehicle terminal's communication module decrypts it to obtain the risk level and related detailed information from the cloud for subsequent steps.
[0066] This invention also provides a method for identifying theft risks of goods, applied to a vehicle-mounted terminal, as described above. Figure 2 , Figure 2 This is the second flowchart illustrating the cargo theft risk identification method provided by the present invention, as follows: Figure 1 As shown, steps 201 to 203 are included: Step 201: Collect image data of the cargo compartment area.
[0067] Specifically, image data refers to visual information captured by cameras installed inside or facing the cargo compartment. This data can take various forms, such as single-frame still images or continuous dynamic video streams. The acquired image data directly reflects the real-time status of the cargo compartment area, including the presence of personnel activity, the condition of the goods, and the operational status of the cameras themselves.
[0068] One specific implementation of this step involves the onboard unit employing a dynamic and differentiated image data acquisition strategy based on the vehicle's operating status. First, the onboard unit uses its integrated GPS module and access to the vehicle's CAN bus to acquire real-time information about the vehicle's driving status, such as determining whether the vehicle is in a "driving state," "idling state," or "stopped state." Then, depending on the different states, the cargo box camera is activated to perform different acquisition actions. For example, in a low-risk driving state, the camera can use a low-power mode, capturing a still image every 30 seconds. In a higher-risk stopped state, it switches to a high-frequency acquisition mode, recording real-time video at a frame rate of 15 frames per second and extracting keyframes periodically.
[0069] Step 202: Analyze the image data to obtain state data.
[0070] After acquiring image data of the cargo compartment area, the vehicle-mounted terminal proceeds to step 202: analyzing the image data to obtain status data. This step is necessary because the raw image data is massive and redundant; directly uploading it all to the cloud would cause a huge network traffic burden and transmission delay. Therefore, the vehicle-mounted terminal needs to utilize its limited computing power to perform a lightweight preliminary analysis and screening of the image data. This step aims to extract core features related to potential risks from the massive amount of raw data and determine whether the anomaly criteria required for reporting to the cloud have been met, thus achieving local data preprocessing.
[0071] In one possible implementation, step 202 specifically includes the following steps: Personnel detection is performed on the image data to obtain the intersection-union ratio (IoU) of the personnel detection bounding box and the coordinates of the cargo compartment warning area; Based on the image data, determine the degree of occlusion and the deflection angle of the camera; Determine whether any one of the intersection-union ratio, occlusion degree, and deflection angle meets the preset conditions. When the preset conditions are met, encapsulate the coordinates of the personnel detection box, occlusion degree, deflection angle, acquisition time, geographical location, and image data into state data.
[0072] Specifically, the first step is to detect people in the image data. A lightweight target detection algorithm, such as a simplified version of YOLO (You Only Look Once) or MobileNet-SSD, runs on the embedded device in the vehicle. This algorithm receives each frame of image captured by the camera as input and quickly outputs the location of all identified personnel targets in the image. The location information is represented by a rectangular personnel detection box, defined by the coordinates (x1, y1, x2, y2) of its upper left and lower right corners. Simultaneously, the vehicle device pre-stores a set of fixed cargo compartment warning area coordinates, which define the boundaries of key areas such as the cargo compartment door. Then, for each detected personnel detection box, the vehicle device calculates the intersection-union ratio (IUU) between it and the rectangular box defined by the preset cargo compartment warning area coordinates. This calculation is performed by dividing the intersection area of the two rectangular boxes by their union area.
[0073] Next, the camera's status is determined. To determine the degree of occlusion, the vehicle-mounted device invokes an image processing algorithm, such as Canny edge detection. This algorithm calculates the total edge length in the current image. By comparing this length with a baseline edge length measured under normal, unobstructed conditions, a percentage of occlusion can be determined. For example, if the current edge length is only 10% of the baseline value, the occlusion level is determined to be 90%. To determine the deflection angle, the vehicle-mounted device reads real-time data from its built-in gyroscope or inertial measurement unit (IMU). This sensor provides the camera's current pitch and yaw angles in space. By comparing these real-time angles with the baseline mounting angles recorded during device initialization, the horizontal and vertical deflection angles can be accurately calculated.
[0074] After all the judgment criteria are calculated, it is determined whether any one of the following—Intersection over Union (IoU), occlusion degree, and deflection angle—meets the preset conditions. The vehicle-mounted terminal internally sets two parallel trigger conditions as preset conditions. The first-level trigger condition is: the IoU of the calculated personnel detection box and the cargo compartment warning area coordinates is greater than 0.3. The second-level trigger condition is: the calculated camera occlusion degree is greater than a certain threshold (e.g., 30%), or its deflection angle is greater than a certain threshold (e.g., 15°). The processing unit on the vehicle-mounted terminal continuously compares the calculated real-time values with these preset thresholds. Once any condition is met, a data encapsulation action is immediately triggered. At this time, the processing unit collects all relevant information about the current event, including the coordinates of the personnel detection box that triggered the conditions, the current occlusion degree and deflection angle values, the geographical location obtained from the GPS module, the acquisition time accurate to the second obtained from the system clock, and the current keyframe image being analyzed. This keyframe image is compressed using an onboard H.265 hardware encoder to reduce its size. Finally, all this information is encapsulated into a structured state data packet according to a predefined format, ready to be sent in the next step.
[0075] Step 203: Send the status data to the cloud so that the cloud can determine the theft risk identification result; the theft risk identification result is to extract multiple first features corresponding to each first person in the cargo compartment area from the status data, and calculate the camera obstruction intention score of each first person based on the multiple first features; determine the second person from the first person based on the camera obstruction intention score; determine the third person who interacts with the second person based on the status data; extract multiple second features corresponding to each third person in the cargo compartment area from the status data, and calculate the theft intention score of each third person based on the multiple second features; and determine the third person based on the theft intention score.
[0076] Specifically, once the status data is encapsulated, the vehicle-mounted communication module immediately performs the transmission operation. This module, such as a 4G or 5G network module, establishes a connection with the cloud platform through a preset server address and port. The communication protocol can adopt the lightweight Message Queuing Telemetry Transport (MQTT) protocol, which has low power consumption and low overhead, making it very suitable for IoT scenarios. The vehicle-mounted device, acting as the publisher, publishes the encapsulated status data packet as the message body (payload) to a specified topic on the cloud server. The cloud platform, acting as the subscriber, receives the message instantly, thus completing the status data transmission process.
[0077] In one possible implementation, the method further includes the following steps: When the degree of occlusion exceeds the first preset threshold or the deflection angle exceeds the second preset threshold, the alarm device on the vehicle is triggered.
[0078] Certain serious equipment anomalies, such as a camera being completely obstructed or its angle being drastically altered, are highly likely to indicate theft. In such emergency situations, waiting for cloud analysis and feedback would result in unacceptable delays. Therefore, it is necessary to implement a low-latency, direct alarm mechanism on the vehicle's system.
[0079] Specifically, the first and second preset thresholds are specific values pre-set in the vehicle's local storage. These thresholds are typically set higher than the thresholds used only to trigger data uploads; they represent a more serious device state, almost certainly indicating malicious activity. For example, the first preset threshold could be set to 80%, indicating that most of the camera's field of view is obstructed. The second preset threshold could be set to 30 degrees, indicating a significant physical change in the camera's position. The alarm device is a physical device connected to the vehicle's main processing unit and may include a high-decibel buzzer and a high-brightness strobe light.
[0080] One specific implementation of this step involves a monitoring program continuously running on the vehicle-mounted processing unit. This program continuously retrieves real-time occlusion levels and deflection angle values from the analysis results of the aforementioned steps. Simultaneously, the program accesses preset first thresholds (e.g., 80%) and second preset thresholds (e.g., 30 degrees) stored locally. An internal logical loop checks in each calculation cycle whether the real-time occlusion level exceeds the first preset threshold or whether the real-time deflection angle exceeds the second preset threshold. Once either of these conditions is met, the processing unit immediately sends a high-level signal to the driver circuit of the external alarm device via a general purpose input / output (GPIO) pin. This signal triggers the driver circuit to power a high-decibel buzzer and strobe light, immediately activating the audible and visual alarm.
[0081] In one possible implementation, the method further includes the following steps: Receive theft risk identification results sent from the cloud; When the theft risk assessment result is high-risk, the vehicle-mounted alarm device is triggered.
[0082] Specifically, the vehicle-mounted communication module continuously listens to a specific network port or subscribes to a specific MQTT topic to prepare for receiving downlink commands from the cloud—the receiving process. When the cloud determines the risk level to be "high risk," it encapsulates a command data packet containing that risk level information and sends it to the vehicle-mounted terminal via the network. After receiving and parsing the data packet, the processing unit checks the risk level field it contains. If the value of this field is confirmed to be "high risk," the processing unit immediately sends a high-level signal to the alarm device's driver circuit via a general-purpose input / output (GPIO) pin, thereby activating the high-decibel buzzer and strobe light. Simultaneously, the vehicle-mounted terminal can also push real-time video captured by the local camera to a designated administrator terminal for remote viewing, based on additional information in the command, such as the push address of the real-time video stream.
[0083] It should be further noted that when the theft risk assessment result is medium risk, the processing unit will not trigger the alarm device. Instead, it will record the event in the local log and initiate a continuous monitoring task. This task will set a timer; every 30 seconds, the timer will trigger the camera to capture an image of the scene and update the image to the cloud via the network module, allowing the cloud to continuously monitor the subsequent development of personnel behavior.
[0084] When the theft risk assessment result is low, the processing unit does not perform any alarm or data upload actions. It only stores the status data related to the event, including keyframe images and analysis parameters, on a local storage device, such as an SD card, for future reference only.
[0085] Reference Figure 3 , Figure 3 This is one of the structural schematic diagrams of the cargo theft risk identification system provided by the present invention. The system is applied in the cloud and includes: The cloud receiving module is used to receive status data sent by the vehicle-mounted terminal. The status data is obtained by analyzing the image data of the cargo compartment area collected by the vehicle-mounted terminal. The cloud processing module is used to extract multiple first features corresponding to each first person in the cargo compartment area from the status data, and calculate the camera occlusion intention score of each first person based on the multiple first features. The cloud processing module is also used to determine the second person from the first person based on the score of each camera's occlusion intention; The cloud processing module is also used to identify third persons who have interacted with the second person based on status data; The cloud processing module is also used to extract multiple second features corresponding to each third person in the cargo compartment area from the status data, and to calculate the theft intent score of each third person based on the multiple second features. The cloud processing module is also used to determine the theft risk identification results based on the scores of each theft intent.
[0086] In one possible implementation, the cloud processing module is further used for: For any first person, extract the speed at which the first person approaches the camera from the state data to obtain the speed feature; The duration during which the first person continuously obstructs the camera is extracted from the state data to obtain the first temporal feature; The distance between the first person and the camera is extracted from the status data to obtain distance features; The first action data of the first person is extracted from the status data to obtain the first action feature; wherein, when the first action data matches the preset action, the value of the first action feature is 1; when the first action data does not match the preset action, the value of the first action feature is 0. Extract time period features from the status data; where the value of the time period feature is 1 when it is in the loading / unloading period and 0 when it is not in the loading / unloading period. The camera occlusion intent score is calculated based on speed characteristics, first time characteristics, distance characteristics, first action characteristics, and time period characteristics.
[0087] In one possible implementation, the cloud processing module is further used for: For any third person, extract the frequency of the third person's contact with the preset item from the state data to obtain frequency characteristics; The probability feature is obtained by extracting the probability of a third person carrying a pre-set theft tool from the state data; The duration of a third person's stay in the cargo compartment's security area is extracted from the status data, and the duration is standardized to obtain a second time feature. Extract the second action of the third person from the state data to obtain the second action feature; Calculate the similarity between the second action feature and the historical theft action features to obtain the theft correlation feature; The theft intent score is calculated based on frequency characteristics, probability characteristics, second time characteristics, and theft correlation characteristics.
[0088] In one possible implementation, the system also includes a cloud sending module for sending the theft risk identification results to the vehicle-mounted terminal.
[0089] Reference Figure 4 , Figure 4 This is the second structural schematic diagram of the cargo theft risk identification system provided by the present invention. The system is applied to a vehicle-mounted terminal and includes: The vehicle-mounted data acquisition module is used to collect image data of the cargo compartment area; The vehicle-mounted processing module is used to analyze image data and obtain status data; The vehicle-mounted sending module is used to send status data to the cloud so that the cloud can determine the theft risk identification result. The theft risk identification result is obtained by extracting multiple first features corresponding to each first person in the cargo compartment area from the status data, and calculating the camera obstruction intention score of each first person based on the multiple first features; determining the second person from the first person based on the camera obstruction intention score; determining the third person who interacts with the second person based on the status data; extracting multiple second features corresponding to each third person in the cargo compartment area from the status data, and calculating the theft intention score of each third person based on the multiple second features; and determining the third person based on the theft intention score.
[0090] In one possible implementation, the vehicle-mounted processing module is further configured to: Personnel detection is performed on the image data to obtain the intersection-union ratio (IoU) of the personnel detection bounding box and the coordinates of the cargo compartment warning area; Based on the image data, determine the degree of occlusion and the deflection angle of the camera; Determine whether any one of the intersection-union ratio, occlusion degree, and deflection angle meets the preset conditions. When the preset conditions are met, encapsulate the coordinates of the personnel detection box, occlusion degree, deflection angle, acquisition time, geographical location, and image data into state data.
[0091] In one possible implementation, the system further includes an on-board alarm module for: When the degree of occlusion exceeds the first preset threshold or the deflection angle exceeds the second preset threshold, the alarm device on the vehicle is triggered.
[0092] In one possible implementation, the system also includes an on-board receiving module for receiving theft risk identification results sent from the cloud; The vehicle-mounted alarm module is also used to trigger the vehicle-mounted alarm device when the theft risk identification result is high-risk.
[0093] It should be noted that the cargo theft risk identification system provided by the present invention can execute the cargo theft risk identification method of any of the above embodiments during specific operation, which will not be elaborated in this embodiment.
[0094] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a cargo theft risk identification method, which includes: receiving status data sent by a vehicle-mounted terminal, the status data being obtained by analyzing image data of the cargo compartment area collected by the vehicle-mounted terminal; extracting multiple first features corresponding to each first person in the cargo compartment area from the status data, and calculating a camera occlusion intention score for each first person based on the multiple first features; determining a second person from the first persons based on the camera occlusion intention scores; determining a third person interacting with the second person based on the status data; extracting multiple second features corresponding to each third person in the cargo compartment area from the status data, and calculating a theft intention score for each third person based on the multiple second features; and determining a theft risk identification result based on the theft intention scores.
[0095] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0096] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer is able to execute the cargo theft risk identification method provided in the above embodiments.
[0097] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the cargo theft risk identification method provided in the above embodiments.
[0098] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0099] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.
[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for identifying the risk of cargo theft, characterized in that, Applied to the cloud, including: Receive status data sent by the vehicle-mounted terminal, wherein the status data is obtained by analyzing image data of the cargo compartment area collected by the vehicle-mounted terminal; Extract multiple first features corresponding to each first person in the cargo compartment area from the state data, and calculate the camera occlusion intention score of each first person based on the multiple first features; Based on the scores for each camera obstruction intent, a second person is determined from the first person; Based on the status data, a third person who interacted with the second person is identified; Extract multiple second features corresponding to each third person in the cargo compartment area from the status data, and calculate the theft intent score of each third person based on the multiple second features; Based on the scores of the stated theft intent, the theft risk identification results are determined.
2. The method for identifying the risk of cargo theft according to claim 1, characterized in that, The step of extracting multiple first features corresponding to each first person in the cargo compartment area from the state data, and calculating the camera occlusion intent score of each first person based on the multiple first features, includes: For any first person, the speed at which the first person approaches the camera is extracted from the state data to obtain a speed feature; The first temporal feature is obtained by extracting the duration during which the first person continuously obstructs the camera from the state data. The distance between the first person and the camera is extracted from the state data to obtain distance features; First action data of the first person is extracted from the state data to obtain a first action feature; wherein, when the first action data matches a preset action, the value of the first action feature is 1; when the first action data does not match the preset action, the value of the first action feature is 0. Extract time period features from the status data; wherein, when it is a loading / unloading period, the value of the time period feature is 1; when it is not a loading / unloading period, the value of the time period feature is 0; The camera occlusion intention score is calculated based on the speed feature, the first time feature, the distance feature, the first action feature, and the time period feature.
3. The method for identifying the risk of cargo theft according to claim 1, characterized in that, The step of extracting multiple second features corresponding to each third person within the cargo compartment area from the status data, and calculating a theft intent score for each third person based on the multiple second features, includes: For any third person, the frequency of the third person's contact with the preset item is extracted from the state data to obtain frequency features; The probability of the third person carrying a preset theft tool is extracted from the state data to obtain a probability feature; The duration of the third person's stay in the cargo compartment warning area is extracted from the status data, and the duration of stay is standardized to obtain a second time feature; Extract the second action of the third person from the state data to obtain the second action feature; Calculate the similarity between the second action feature and the historical theft action features to obtain the theft correlation feature; The theft intent score is calculated based on the frequency feature, the probability feature, the second time feature, and the theft correlation feature.
4. The method for identifying the risk of cargo theft according to claim 1, characterized in that, Also includes: The theft risk identification results are sent to the vehicle-mounted terminal.
5. A method for identifying the risk of cargo theft, characterized in that, Applications in automotive applications include: Collect image data of the cargo compartment area; The image data is analyzed to obtain status data; The status data is sent to the cloud so that the cloud can determine the theft risk identification result; the theft risk identification result is obtained by extracting multiple first features corresponding to each first person in the cargo compartment area from the status data, and calculating the camera obstruction intention score of each first person based on the multiple first features; determining a second person from the first persons based on the camera obstruction intention score; determining a third person who interacts with the second person based on the status data; extracting multiple second features corresponding to each third person in the cargo compartment area from the status data, and calculating the theft intention score of each third person based on the multiple second features; and determining the third person based on the theft intention score.
6. The method for identifying the risk of cargo theft according to claim 5, characterized in that, The analysis of the image data to obtain state data includes: Perform personnel detection on the image data to obtain the intersection-union ratio (IoU) of the personnel detection bounding box and the cargo compartment warning area coordinates; Based on the image data, determine the degree of occlusion and the deflection angle of the camera; Determine whether any one of the intersection-union ratio, the degree of occlusion, and the deflection angle meets a preset condition, and when the preset condition is met, encapsulate the coordinates of the personnel detection box, the degree of occlusion, the deflection angle, the acquisition time, the geographical location, and the image data into the state data.
7. The method for identifying the risk of cargo theft according to claim 6, characterized in that, Also includes: When the degree of occlusion is greater than a first preset threshold or the deflection angle is greater than a second preset threshold, the alarm device on the vehicle is triggered.
8. The method for identifying the risk of cargo theft according to claim 5, characterized in that, Also includes: Receive the theft risk identification result sent by the cloud; When the theft risk identification result is high-risk, the alarm device on the vehicle is triggered.
9. A cargo theft risk identification system, characterized in that, Applied to the cloud, including: The cloud receiving module is used to receive status data sent by the vehicle-mounted terminal. The status data is obtained by analyzing the image data of the cargo compartment area collected by the vehicle-mounted terminal. The cloud processing module is used to extract multiple first features corresponding to each first person in the cargo compartment area from the status data, and calculate the camera occlusion intention score of each first person based on the multiple first features. The cloud processing module is also used to determine the second person from the first person based on the occlusion intention scores of each of the cameras; The cloud processing module is also used to determine, based on the status data, a third person who has interacted with the second person; The cloud processing module is also used to extract multiple second features corresponding to each of the third persons in the cargo compartment area from the status data, and to calculate the theft intent score of each of the third persons based on the multiple second features; The cloud processing module is also used to determine the theft risk identification result based on the theft intent scores of each item.
10. A cargo theft risk identification system, characterized in that, Applications in automotive applications include: The vehicle-mounted data acquisition module is used to collect image data of the cargo compartment area; The vehicle-mounted processing module is used to analyze the image data to obtain status data; The vehicle-mounted sending module is used to send the status data to the cloud so that the cloud can determine the theft risk identification result; the theft risk identification result is obtained by extracting multiple first features corresponding to each first person in the cargo compartment area from the status data, and calculating the camera obstruction intention score of each first person based on the multiple first features; determining a second person from the first persons based on the camera obstruction intention score; determining a third person who interacts with the second person based on the status data; extracting multiple second features corresponding to each third person in the cargo compartment area from the status data, and calculating the theft intention score of each third person based on the multiple second features; and determining the theft risk identification result based on the theft intention score.