An automatic alignment vehicle method and an autonomous navigation bulk inspection vehicle method for an under-vehicle inspection robot

Abstract

The application discloses an automatic alignment vehicle method and an autonomous navigation batch vehicle inspection method of a vehicle bottom security inspection robot. Through combination of a front camera real-time video stream and a target detection algorithm, a vehicle to be detected is accurately identified, and a target vehicle is selected through an app end. A position of the target vehicle in an image is automatically calculated and converted into a laser radar coordinate, and accurate scanning of a vehicle contour is realized. The robot speculates a vehicle position according to a laser radar scanning result and automatically aligns, and starts a vehicle bottom scanning task. On a user end, a parking lot model can be constructed through an app. Optional detection modes include a single vehicle mode, a longitudinal vehicle mode, a transverse vehicle mode and a multi-row custom mode. According to different modes selected by a user, the vehicle bottom security inspection robot performs corresponding autonomous navigation batch vehicle inspection operations, so as to meet diversified security inspection requirements.

Description

TECHNICAL FIELD

[0001] The present application relates to security detection technology, in particular to a method for automatically aligning a vehicle and a method for autonomously navigating batch inspection of vehicles. BACKGROUND

[0002] The vehicle bottom security inspection robot is an automated device specially used for vehicle bottom security inspection; it is usually equipped with high-definition cameras, sensors, lighting equipment, etc., and can comprehensively and meticulously scan and inspect the vehicle bottom without direct human contact with the vehicle. The vehicle bottom security inspection robot replaces manual inspection of the vehicle bottom, can quickly complete vehicle bottom inspection, improves the efficiency of security inspection, and reduces vehicle waiting time. Through advanced image recognition and analysis technology, it can accurately detect abnormal conditions of the vehicle bottom, such as hidden items and modified parts. It avoids direct contact of security personnel with the potentially dangerous vehicle bottom, ensuring the safety of security personnel. It can autonomously operate according to pre-set programs and paths, reducing human operation errors.

[0003] However, the current vehicle bottom security inspection robot may have some deficiencies, such as being unable to autonomously implement vehicle alignment detection and return, requiring manual operation; after inspecting one vehicle, it needs to be manually moved to another adjacent vehicle, and its autonomous batch inspection capability is limited. The navigation strategy of the robot is not strong or flexible enough, and it cannot autonomously move to adjacent vehicles to achieve batch inspection. It cannot automatically detect multiple rows of vehicles, etc., which to some extent limits its widespread application and performance.

[0004] The existence of the above problems leads to many deficiencies in the use of the existing vehicle bottom security inspection robot:

[0005] Low security inspection efficiency: After each vehicle inspection, the robot needs to be manually moved, which consumes a lot of time and reduces the overall security inspection speed, especially in places with high vehicle flow, which can easily cause congestion.

[0006] Increased labor costs: Relying on manual movement of the robot increases labor input and operating costs.

[0007] Increased safety risks: Due to the inability to automatically detect multiple rows of vehicles, some vehicle bottom security inspections may be missed, posing a safety risk. In addition, due to the inability to automatically align the center of the vehicle to be inspected, the chassis image may not be fully captured.

[0008] Poor work continuity: The inability to map the parking lot in advance makes the robot's work lack planning and coherence, affecting the overall security inspection effect.

[0009] Decreased service quality: The security inspection process is not smooth and convenient, which can affect user experience and reduce the quality and satisfaction of related services.

[0010] Weak emergency response capability: in the emergency situation that needs to quickly check a large number of vehicles, such as major activities or emergencies, it is difficult to complete the task efficiently. SUMMARY

[0011] In view of the deficiencies of the prior art, the present application aims to provide an automatic alignment vehicle method of a vehicle bottom inspection robot and an autonomous navigation batch inspection vehicle method, which can automatically align the center of the vehicle to be tested and realize batch scanning inspection of the vehicle chassis.

[0012] The present application is realized by the following technical solutions:

[0013] An automatic alignment vehicle method of a vehicle bottom inspection robot, comprising the following steps:

[0014] Step one: after the robot starts, real-time video stream from the front camera is received, and image processing is performed internally, target detection algorithm is run on the obtained data, and all vehicles in the image are identified, and each identified vehicle is marked with a rectangular frame (x1, y1, x2, y2);

[0015] Step two: the video stream data of the front camera is transmitted to the app end through a wireless network;

[0016] Step three: after the user clicks the vehicle to be tested in the picture on the app display interface, the click coordinates c1 are sent to the robot, and according to the click coordinates c1 of the user, the robot program determines which rectangular frame is the vehicle selected by the operator, then calculates the horizontal position range of the vehicle in the image, represented as c2 = (xl, xr), that is, the horizontal coordinates of the left and right edges of the vehicle in the image;

[0017] Step four: convert the c2 value in the image coordinate system to the actual physical coordinate system of the laser radar, after the coordinate conversion, calculate the corresponding angle interval (a1, a2) that the laser radar should scan;

[0018] Step five: the laser radar scans according to the starting angle a1 and the ending angle a2 to search for the left wheel and the right wheel of the vehicle, and obtains the coordinate positions c3, c4; the positions c3, c4 are used to infer the position of the vehicle to be tested, and after obtaining the position of the vehicle to be tested, automatic alignment is performed, and the vehicle bottom scanning task is started.

[0019] Further, the automatic alignment process of step five is:

[0020] Step one: when the robot receives a double-click instruction, it enters the automatic alignment vehicle mode;

[0021] Step two: the robot scans the current vehicle front wheel track through the laser radar, selects the double-wheel line axis to the robot side 0.5m distance to find the target point;

[0022] Step three: adjust the robot car head extension line to align the target point, and proceed to the target point;

[0023] Step four: adjust the robot angle in place to make its car body completely aligned with the target path midline.

[0024] A kind of vehicle bottom security robot autonomous navigation batch checks vehicle method, constructs parking lot model and detection model by user end app, detection mode includes longitudinal train mode, horizontal row car mode and multiple row custom mode;

[0025] Longitudinal train mode:

[0026] S1.1, user input longitudinal train vehicle total number, app operation moves vehicle bottom security robot to the first vehicle head position of longitudinal train, when the front camera recognizes the first vehicle to be detected by target recognition algorithm, the laser radar is aligned to the first vehicle to scan the front wheel track, and the first target point is found by selecting the double-wheel line axis to the robot side 0.5m distance;

[0027] S1.2, vehicle bottom security robot reaches the first target point, so that the laser reaches both sides wheel track is same, the license plate information is identified by the front camera, and the robot proceeds to pass through the first vehicle bottom, when the vehicle top distance measuring assembly detects that a vehicle detection is completed according to distance change, the complete vehicle bottom image collected by the linear array camera assembly is acquired, and the parking instruction robot stops;

[0028] S1.3, the total number of longitudinal train vehicles is reduced by one, and the laser radar is aligned to the next vehicle by similar method, and the license plate information is identified. Whether the license plate is found or not, the vehicle bottom security robot scans the wheel track in front of the vehicle through the laser radar, and the vehicle is calibrated again. After calibration, the next vehicle is detected, and the complete vehicle bottom image of each vehicle is obtained until all vehicle bottom detection is completed.

[0029] Horizontal row car mode:

[0030] S2.1, app operation moves vehicle bottom security robot to the first vehicle head position of longitudinal train, and the robot front camera recognizes the vehicle to be detected by target recognition algorithm;

[0031] S2.2, laser radar scans the front wheel track, and selects the double-wheel line axis to the robot side 3m distance to find the first target point;

[0032] S2.3, autonomously navigate to the first target point, rotate the robot in place so that the laser radar has the same wheelbase on both sides, at this time the robot is aligned with the first vehicle, and the license plate is recognized; the robot moves forward and passes through the bottom of the first vehicle, when the top ranging component determines that the robot has completely passed through the bottom of the vehicle, the system obtains a complete vehicle bottom picture, and then retreats along the original route to return;

[0033] S2.4, the robot returns to the posture of aligning with the first vehicle at this time, and determines whether all vehicle detection is completed, if completed, jump to step S2.6, if not completed, execute step S2.5;

[0034] S2.5, rotate by 90 degrees to the side of the next vehicle, and then drive forward by a fixed distance D, wherein the fixed distance D is calculated by the formula , wherein W represents the width of the parking space, and n represents the number of empty parking spaces, after reaching the target point, immediately rotate by 90 degrees in the opposite direction, and cycle steps S2.1 to S2.4;

[0035] S2.6, end the task, and output and store all vehicle license plate and complete chassis picture information;

[0036] Multi-row self-defined mode:

[0037] In the multi-row mode, one of the vertical column vehicle mode or the horizontal row vehicle mode is selected to execute the single-row mode, after executing the single-row mode, 180 degrees are rotated to complete scanning of the vehicles on the opposite side in the reverse direction, the task is ended, and all vehicle license plate and complete chassis picture information are output and stored.

[0038] Further, in the horizontal row vehicle mode, the user first inputs the number of horizontal row parking spaces and the width of the parking space information on the app side, marks the empty parking spaces, and then generates a horizontal row vehicle batch inspection model.

[0039] The present application has the following beneficial effects:

[0040] (1) By combining the real-time video stream of the front camera with the target detection algorithm, the vehicle to be detected is accurately recognized, and the target vehicle is selected through the app side, the position of the target vehicle in the image is automatically calculated and converted into laser radar coordinates, and accurate scanning of the vehicle contour is realized. The robot predicts the position of the vehicle according to the scanning result of the laser radar and automatically aligns, and starts the vehicle bottom scanning task, thereby greatly improving the automation, accuracy and efficiency of the vehicle bottom security check, reducing manual operation, and improving work convenience and safety.

[0041] (2) Batch scanning and inspection of vehicle chassis can be realized, avoiding the need for manual movement to another adjacent vehicle after each vehicle is inspected;

[0042] (3) In the application scene of vehicle chassis security inspection, most of them do not have the function of building a slam map in advance for the parking lot, and the application only needs the user to input a small amount of parameters to realize the batch detection of the mainstream standard parking lot vehicles, thereby improving the work efficiency.

[0043] (4) In the technical solution, the alignment algorithm of a single vehicle is corrected according to the front double-wheel track, which can accurately align the vehicle, and the algorithm is used to correct the new vehicle each time the scanning is performed, so that the cumulative error is avoided, and the task success rate of the self-batch vehicle detection of the system is ensured.

[0044] (5) The application has expansibility, and the inspection of multi-row horizontal vehicles can be realized by the simple combination task of the application end, and for most standard parking lots, the scanning of all vehicles in the parking lot can be completed semi-autonomously without mapping. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 is a structure diagram of a vehicle bottom security inspection robot;

[0046] Figure 2 is a longitudinal vehicle mode schematic diagram;

[0047] Figure 3 is a longitudinal vehicle mode detection algorithm flowchart;

[0048] Figure 4 is a horizontal row vehicle mode schematic diagram;

[0049] Figure 5 is a horizontal row vehicle mode detection algorithm flowchart;

[0050] Figure 6 is a multi-row vehicle mode schematic diagram.

[0051] In the figure: 1-linear array camera assembly, 2-top distance measuring assembly, 3-top wide-angle camera, 4-laser radar, 5-front camera. DETAILED DESCRIPTION

[0052] The application will be further described in detail below in combination with specific embodiments, which are an explanation of the application rather than a limitation.

[0053] For example, Figure 1As shown, the present application provides a vehicle chassis inspection robot, which comprises a carrying trolley and a linear array camera assembly 1, a top ranging assembly 2, a top wide-angle camera 3, a laser radar 4, a front camera 5, a main control board and a data transmission module installed on the carrying trolley. The vehicle chassis inspection robot combines linear array cameras, cameras, laser radars and other sensors, and through laser radar and camera fusion algorithms, it can automatically align the vehicle, enter the bottom of the vehicle and check whether there is illegal hiding or modified security equipment in the vehicle chassis. The carrying trolley adopts a four-wheel differential model chassis, which can realize forward movement, backward movement and differential rotation at any angle in place.

[0054] Through the linear array camera, the camera, the laser radar and other sensors, the chassis image and video data are collected, a large amount of collected data is compressed and encoded, and the encoded data is transmitted to the user terminal app through the Wi-Fi data transmission module.

[0055] The carrying trolley driving system uses a motor driver or other power control equipment to convert the planned motion instructions into motor speed, steering and other control signals, so as to realize the forward movement, backward movement, turning and other actions of the vehicle chassis inspection robot. The position, speed and attitude information of the vehicle chassis inspection robot are obtained in real time by using encoders, gyroscopes and other sensors, compared with the planned path, closed-loop controlled, and the motion is adjusted in time to ensure the accuracy and stability of the driving.

[0056] The data returned by the user terminal app is decompressed, decoded and analyzed, and useful information is extracted.

[0057] In order to realize more efficient vehicle chassis inspection, the present application further provides an automatic vehicle alignment method of the vehicle chassis inspection robot, which can automatically identify and locate the vehicle to be inspected after starting, so as to ensure that the robot can accurately align the chassis of the target vehicle and thus perform effective inspection.

[0058] The automatic vehicle alignment method of the vehicle chassis inspection robot of the present application comprises the following steps:

[0059] Step one: after the robot starts, the real-time video stream from the front camera 5 is received, and image processing is performed inside, and the target detection algorithm is run on the obtained data to identify all vehicles in the image, and each identified vehicle is marked with a rectangular frame (x1, y1, x2, y2);

[0060] Step two: the video stream data of the front camera 5 is transmitted to the app terminal through a wireless network;

[0061] Step three: the user clicks the vehicle to be tested in the picture on the app display interface, and clicks the coordinate c1 to send to the robot. According to the user's click coordinate c1, the robot program determines which rectangular frame is the vehicle selected by the operator, and then calculates the lateral position range of the vehicle in the image, represented as c2 = (xl, xr), that is, the horizontal coordinates of the left and right edges of the vehicle on the image.

[0062] Step four: convert the c2 value in the image coordinate system to the actual physical coordinate system of the laser radar 4. After the coordinate conversion, the corresponding laser radar 4 scanning angle (a1, a2) is calculated.

[0063] Step five: the laser radar 4 scans according to the starting angle a1 and the ending angle a2 to search for the left wheel and the right wheel of the vehicle, and obtains the coordinate positions c3 and c4. The position of the vehicle to be tested is inferred based on the positions c3 and c4, and after obtaining the position of the vehicle to be tested, automatic alignment is performed to start the vehicle bottom scanning task.

[0064] The flow of automatic alignment is as follows:

[0065] Step one: when the robot receives a double-click instruction, it enters the automatic alignment vehicle mode.

[0066] Step two: the robot scans the current vehicle front wheel track through the laser radar 4, and selects the target point 0.5m away from the robot side along the axis of the double-wheel connecting line.

[0067] Step three: adjust the robot car head extension line to align the target point and proceed to the target point.

[0068] Step four: adjust the robot angle in place to make the robot body completely aligned with the center perpendicular line of the target path.

[0069] The user end constructs the parking lot model through the app operation. The user's optional detection mode includes: single vehicle mode, longitudinal vehicle mode, transverse vehicle mode and multiple row custom mode.

[0070] The navigation method of the vehicle bottom security robot for self-batch inspection of parking lot vehicles of the present application includes the following contents:

[0071] (1) Longitudinal vehicle mode:

[0072] As shown in Figure 2 , the operation app makes the vehicle bottom security robot pass through all the longitudinally arranged vehicles in sequence, obtains the image collected by the line array camera assembly 1 according to the distance change of the robot and the vehicle bottom obtained by the robot top distance measuring assembly 2, and sequentially obtains the vehicle bottom picture grouping, storage and output.

[0073] As shown in Figure 3 , the specific batch inspection vehicle method of the longitudinal vehicle mode is:

[0074] 1) The user inputs the total number of longitudinal train vehicles, and the operation app moves the train bottom security robot to the 45-degree angle range perpendicular to the head of the first vehicle of the longitudinal train. When the front camera 5 identifies the first vehicle to be detected through a target recognition algorithm, the laser radar 4 is aligned to scan the front wheel track of the first vehicle, and the first target point is found at a distance of 0.5 m from the robot side along the central axis of the double-wheel connecting line.

[0075] 2) The train bottom security robot reaches the first target point, so that the laser radar reaches both sides of the wheel track. The front camera 5 identifies the license plate information, and the robot moves forward through the bottom of the first vehicle. When the vehicle top distance measuring assembly 2 detects that a vehicle has been detected according to the distance change, the complete vehicle bottom image collected by the linear camera assembly 1 is obtained, and the stop command is issued to stop the robot.

[0076] 3) The total number of longitudinal train vehicles is reduced by one, and the laser radar 4 is aligned to the next vehicle through a similar method. Whether the license plate is found or not, the train bottom security robot scans the wheel track in front through the laser radar 4, and the vehicle is calibrated again. After calibration, the next vehicle is detected until all vehicle bottom images are obtained.

[0077] (2) Horizontal row car mode:

[0078] As shown in Figure 4 , the user constructs a batch inspection model of horizontal row cars on the app side, and the user moves the train bottom security robot to the 45-degree angle range perpendicular to the head of the first vehicle through the app, and then a horizontal row car inspection task can be started. Each vehicle is detected in turn until all vehicle bottom images are obtained.

[0079] As shown in Figure 5 , the user needs to input the width of the horizontal row parking space and the information whether the horizontal row inspection vehicle is parked full. On the app side, the user can construct a batch inspection model of horizontal row cars. The user first inputs the number of horizontal row parking spaces and the width of the parking space, and the application side displays the corresponding horizontal row parking lot model. Then the user manually marks the empty parking space, and the application side automatically generates a task model.

[0080] 1) The robot front camera 5 identifies the vehicle to be detected through a target recognition algorithm;

[0081] 2) The laser radar scans the front wheel track, and the first target point is found at a distance of 3 m from the robot side along the central axis of the double-wheel connecting line.

[0082] 3) Autonomous navigation to the first target point, rotate the robot in place, so that the laser reaches both sides of the wheelbase is the same, at this time the robot aligns the first vehicle, license plate recognition; The robot moves forward, through the first car under the car, when the top ranging components determine that the robot has completely passed through the car bottom, the system obtains the complete car bottom picture, then back the way back;

[0083] 4) The robot returns to the first vehicle alignment posture, determine whether all vehicle detection is completed, if completed jump to step 6), if not completed execute step 5);

[0084] 5) Turn 90 degrees to the side of the next vehicle, then drive a fixed distance D, where the fixed distance D is calculated by the formula , where W represents the width of the parking space, n represents the number of empty parking spaces, after reaching the target point, immediately rotate 90 degrees in the opposite direction, cycle steps 1) to 4);

[0085] 6) End of task, output storage of all vehicle license plate and complete chassis picture information.

[0086] (3) Multi-row custom mode:

[0087] As shown in Figure 6 , the user can select multi-row mode on the application side, multi-row mode selects one of the vertical column car mode or horizontal car mode to execute single row mode, after executing single row, rotate 180 degrees, complete the opposite side of the vehicle scanning in reverse, end the task, output storage of all vehicle license plate and complete chassis picture information.

Claims

1. A method for autonomously navigating batch inspection of vehicles using an under-vehicle security inspection robot, characterized in that: Build parking lot models and detection models through the user-side app. Detection modes include longitudinal train mode, transverse car mode, and multi-row custom mode. Longitudinal train mode: S1.

1. The user enters the total number of vehicles in the train and uses the app to move the undercarriage inspection robot to the front of the first vehicle in the train. When the front camera identifies the first vehicle to be inspected using the target recognition algorithm, it uses the LiDAR to scan the front wheel track of the first vehicle and locates the first target point 0.5m to the robot's side, along the center axis of the two wheels. S1.

2. The under-vehicle inspection robot reaches the first target point, aligning the wheelbases of both sides with the laser radar. The robot uses the front-facing camera to identify the license plate information and moves forward, crossing under the first vehicle. When the top-mounted ranging component detects a vehicle based on the change in distance, it acquires a complete under-vehicle image captured by the linear array camera component and issues a stop command to the robot. S1.

3. The total number of vehicles in the longitudinal train is reduced by one. The LiDAR is then aligned with the next vehicle using a similar method to identify the license plate. Regardless of whether the license plate is found, the undercarriage inspection robot scans the wheelbase ahead with the LiDAR and recalibrates the vehicle. After calibration, the robot begins inspecting the next vehicle, continuing until all vehicles have been fully inspected, obtaining undercarriage images of each vehicle. Horizontal car mode: S2.

1. Use the app to move the undercarriage inspection robot to the front of the first car in the longitudinal train. The robot's front camera (5) identifies the vehicle to be inspected through the target recognition algorithm. S2.

2. Scan the front wheel track with the LiDAR and locate the first target point 3 meters to the robot's side, along the center axis connecting the two wheels. S2.

3. Autonomously navigate to the first target point and rotate the robot in place so that the wheelbase of both wheels is the same as the laser radar. The robot then aims at the first vehicle and performs license plate recognition. The robot moves forward and passes under the first vehicle. When the top ranging component determines that the robot has completely passed under the vehicle, the system obtains a complete image of the vehicle's underside and then reverses to return home. S2.4: The robot returns to the position aimed at the first vehicle and determines whether all vehicles have been inspected. If so, it proceeds to step S2.6; if not, it proceeds to step S2.

5. S2.

5. Turn 90 degrees toward the next vehicle, then drive forward a fixed distance D, calculated by the formula D = W * (n + 1), where W is the width of the parking space and n is the number of empty spaces. Upon reaching the target point, immediately turn 90 degrees in the opposite direction, repeating steps S2.1 through S2.

4. S2.

6. End the task and output and store all vehicle license plates and complete chassis image information; Multi-row custom mode: When executing the multi-row mode, select either the longitudinal train mode or the transverse vehicle mode to execute the single-row mode. After completing the single-row scan, rotate 180 degrees and complete the scan of the opposite row of vehicles in reverse. End the task and output and store the license plates and complete chassis images of all vehicles. The longitudinal train mode, the transverse train mode and the multi-row custom mode adopt an automatic vehicle alignment method, including the following steps: Step 1: After the robot starts, it receives a real-time video stream from the front camera and performs image processing internally. It then runs an object detection algorithm on the acquired data to identify all vehicles in the image. Each identified vehicle is marked with a rectangular box (x1, y1, x2, y2). Step 2: The video stream data of the front camera is transmitted to the app via a wireless network; Step 3: After the user clicks on the vehicle to be tested on the app's display interface, the click coordinates c1 are sent to the robot. Based on the user's click coordinates c1, the robot program determines which rectangular box represents the vehicle selected by the operator and then calculates the lateral position range of the vehicle in the image, expressed as c2 = (xl, xr), which are the horizontal coordinates of the left and right edges of the vehicle on the image. Step 4: Convert the c2 value in the image coordinate system to the actual physical coordinate system of the laser radar. After completing the coordinate conversion, calculate the corresponding angle interval (a1, a2) that the laser radar should scan. Step 5: The laser radar scans based on the starting angle a1 and the ending angle a2, searches for the left and right wheels of the vehicle, and obtains their coordinate positions c3 and c4; the position of the vehicle to be tested is inferred from the positions c3 and c4, and after obtaining the position of the vehicle to be tested, automatic alignment is performed to start the under-vehicle scanning task.

2. The autonomous navigation batch inspection vehicle method according to claim 1, characterized in that: The process of automatic alignment described in step 5 is as follows: Step 1: When the robot receives the double-click command, it enters the automatic vehicle alignment mode; Step 2: The robot scans the front wheel track of the current vehicle using the laser radar, and selects the target point 0.5m away from the center axis of the two-wheel connection to the robot side; Step 3: Adjust the extension line of the robot's head to align with the target point and move to the target point; Step 4: Adjust the robot's angle in situ so that its body is completely aligned with the mid-perpendicular line of the target path.

3. The autonomous navigation batch vehicle inspection method according to claim 1, characterized in that: In the horizontal parking mode, the user first enters the number of horizontal parking spaces, parking space width information, marks empty parking spaces, and then generates a horizontal parking batch inspection model.

Images