Damage detection apparatus and method

Abstract

The invention provides an apparatus and method for automatically detecting damage to a vehicle which minimizes the time taken to detect damage to each vehicle. Preferably, the act of damage detection is also carried out in conjunction with the act of repairing any detected damage, where appropriate. To this end, the invention provides a damage detection unit configured to detect damage on a vehicle. The damage detection unit comprises a receiving unit configured to receive images from a camera, and a determining unit configured to determine damage on the vehicle based on the received images. Furthermore, the damage detection unit further comprises a storing unit configured to store information about previously determined damage present on the vehicle, and a comparing unit configured to compare the damage determined by the determining unit with the information about the previous damage stored by the storing unit.

Description

[0001] This application claims priority to UK Patent Application No. GB1909578.5, filed on 3 July 2019, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This invention generally relates to the field of damage detection, and more specifically, to apparatus and methods for detecting damage on vehicles. Background Technology

[0003] Many companies use fleets to accomplish specific tasks. For example, delivery vehicles are typically owned by companies that transport goods to customers and / or other companies. Similarly, car rental companies have the ability to provide rental cars to customers and / or other companies for a specific period of time. Managing and maintaining a fleet is a challenge for these companies.

[0004] One specific detail that must be maintained is the mechanical and electrical condition of the vehicles. Companies managing the fleet need to keep the vehicles in an operational and undamaged condition. Furthermore, if damage is caused to a vehicle, the driver responsible must be properly held accountable, and the driver should be retrained as appropriate.

[0005] Furthermore, damage often impairs the vehicle's mechanical and electrical functions. For example, damage to the tire sidewalls while the vehicle is in use can cause premature tire failure, rendering the vehicle inoperable. Such malfunctions will prevent companies from using the vehicle to transport goods and will prevent rental companies from leasing the vehicle until it is repaired.

[0006] Similarly, damage to vehicle lights (headlights, turn signals, brake lights) would also prevent a company from operating such damaged vehicles, as these damaged vehicles could be at risk of colliding with other vehicles due to the broken lights (e.g., at night).

[0007] However, due to the time spent on each vehicle and the number of vehicles that need to be inspected, it is impractical to inspect the entire fleet after each use to ensure the mechanical and electrical condition of the vehicles.

[0008] Therefore, there is a need to provide a system that allows for the rapid and accurate identification of mechanical and electrical damage to a vehicle, and, if necessary, the correction of such damage. Summary of the Invention

[0009] To address the aforementioned problems, the present invention aims to provide an apparatus and method for automatically detecting vehicle damage, which minimizes the time spent detecting damage to each vehicle. Preferably, where appropriate, the damage detection is performed concurrently with the repair of any detected damage.

[0010] Generally speaking, the present invention introduces a vehicle damage detection unit configured to detect damage on a vehicle based on an image of the vehicle and compare the detected damage with previous damage to determine whether new damage has occurred.

[0011] According to the present invention, a damage detection unit is provided, configured to detect damage on a vehicle. The damage detection unit includes a receiving unit configured to receive images from a camera, and a determining unit configured to determine damage on the vehicle based on the received images. Furthermore, the damage detection unit further includes a storage unit configured to store information about previously determined damage on the vehicle, and a comparison unit configured to compare the damage determined by the determining unit with the information about the previous damage stored in the storage unit.

[0012] The present invention further provides a damage detection system, which includes a camera configured to capture images of a vehicle and a damage detection unit as described above.

[0013] The present invention further provides a damage detection method for detecting damage on a vehicle. The method includes the following steps: receiving an image from a camera; determining damage on the vehicle based on the received image; retrieving information about previously determined damage on the vehicle; and comparing the damage determined in the determining step with the information about the previously determined damage retrieved in the retrieving step. Attached Figure Description

[0014] Specific embodiments of the invention are described by way of example only with reference to the accompanying drawings, wherein the same reference numerals denote the same or corresponding parts, and wherein:

[0015] Figure 1 A schematic diagram of a damage detection unit, a camera, and a directional unit according to a first embodiment of the present invention is depicted.

[0016] Figure 2 A potential layout for cameras configured as imaging vehicles is depicted;

[0017] Figure 3 A schematic diagram of a damage detection unit according to a first embodiment of the present invention is depicted;

[0018] Figure 4 Images of the vehicle captured by a camera and areas of interest on the vehicle are depicted;

[0019] Figure 5 A method for detecting damage on a vehicle according to a first embodiment of the present invention is described;

[0020] Figure 6 Illustrative diagrams depict general-purpose computer hardware and software implementations that provide certain aspects, as specifically described in the specification. Detailed Implementation

[0021] First Specific Implementation

[0022] Figure 1 A damage determination unit 400 according to a first embodiment of the present invention is shown. Figure 1 Additional devices that can be used in conjunction with the damage determination unit 400 of the first specific embodiment are also shown.

[0023] Specifically, Figure 1 Cameras 201-204 configured as imaging vehicles 100 are shown. Figure 1 The image shows vehicle 100 with different types of damage, such as dents 101, scratches 102, and missing parts 103. For example, missing parts 103 may include missing mechanical or electrical components, such as bumpers, headlights, turn signals, brake lights, etc. Similarly, scratches 102 may include damage to the vehicle's tires by scratching, which may lead to premature tire failure. Dents may include damage to the vehicle body from a collision.

[0024] Furthermore, a direction unit 205 is shown, configured to direct vehicles along a normal route, for example, to refuel, refill cargo, or transfer to a new customer. In this example, the direction unit 205 is shown in the form of a traffic light; however, other forms are conceivable.

[0025] Alternatively, the orientation unit 205 can direct vehicle 100 to a workshop or other facility to correct existing damage on the vehicle. In this regard, the damage determination unit 400 can be further configured to automatically schedule vehicle 100 for repair in the workshop. Alternatively, the orientation unit 205 can be configured to notify vehicle 100 to return to the workshop at a future time. For example, the workshop may currently be full of vehicles under repair. Therefore, by scheduling vehicle 100 to enter the workshop at a future time and notifying vehicle 100 to return at that time, the vehicle can be monitored even when the workshop is full and vehicle 100 does not require immediate repair. In this way, automated vehicle repair scheduling can be achieved.

[0026] If vehicle 100 is an autonomous vehicle (e.g., a vehicle that moves without the direct assistance of any driver / passenger), then orientation unit 205 can command the autonomous vehicle to proceed to the location of the damage identified by workshop repair damage determination unit 400. Therefore, orientation unit 205 is conceived as a unit configured to command the autonomous vehicle.

[0027] although Figure 1Four cameras 201-204 are depicted, but only one camera 201 is required for use with the damage determination unit 400. The other three cameras 202-204 are optional. Furthermore, although four cameras 201-204 are shown, it is conceivable that more than four cameras could be used to determine damage on the vehicle 100.

[0028] Cameras 201-204 can be configured to surround vehicle 100 to capture images of vehicle 100 from different angles / positions. A further camera can be positioned to capture images of specific areas of vehicle 100 most vulnerable to damage.

[0029] The damage determination unit 400 is configured to receive images of the vehicle 100 from camera 201 (and optionally, any one of cameras 202-204), and is configured to determine whether new, previously undetected damage has occurred on the vehicle 100 and / or whether the damage has worsened since the last determination. Based on the determined damage, the damage determination unit 400 may be configured to control the orientation unit 205. The orientation unit 205 controls the vehicle 100 to perform one of several actions. For example, the orientation unit 205 may be configured to instruct the vehicle to proceed without any remedial action on the detected damage. Alternatively, the vehicle 100 may be instructed to proceed to a workshop, etc., where the detected damage can be corrected.

[0030] Figure 2 An example layout of four cameras 201-204 around a vehicle is depicted. It can be seen that by placing a camera primarily capturing images of the corners of vehicle 100, the damage determination unit 400 can provide the most accurate damage determination, as the damage is most likely to occur in the corners of vehicle 100. Additional cameras can be used; for example, a camera can be further mounted to capture images of the roof of vehicle 100. Similarly, a camera can be included to capture images of the license plate number / plate of vehicle 100. In this way, damage to specific areas of vehicle 100 can be captured.

[0031] It can be observed that using cameras that capture high-quality images results in more accurate damage detection, both in terms of the type and location of the damage identified. Furthermore, compared to low-quality images, cameras capturing higher-quality images can capture more instances of damage on vehicles 100.

[0032] It was also found that using consistent lighting on the vehicle improved the results. In other words, even at night or in adverse weather conditions, illuminating the vehicle with consistent lighting makes damage detection more accurate. In this way, regardless of the time of day or year when images are taken, the images captured by the camera have 100% consistent lighting on the vehicle.

[0033] It can also be observed that capturing high dynamic range (HDR) images of vehicle 100 further improves damage detection. In this regard, camera 201 can be configured to capture several images at various exposure values. The several images are combined into a single image, thus possessing enhanced dynamic range characteristics.

[0034] Figure 3 A damage detection unit 400 according to a first embodiment of the present invention is depicted. The damage determination unit 400 includes a receiving unit 401, a determining unit 402, a storage unit 405, and a comparison unit 404. Optionally, the damage determination unit 400 may further include at least one of a training unit 403 and a command unit 406.

[0035] The receiving unit 401 is configured to receive images of the vehicle from the camera 201. Optionally, the receiving unit 401 may be further configured to process the images such that all images processed by the determination unit 402 have similar compositions. For example, the receiving unit 401 may be further configured to crop the received images, adjust the brightness of the images, balance the colors, or remove the background of the images, thereby leaving only the vehicle 100 in the images. In this way, consistent performance can be achieved because the images do not change significantly between shots.

[0036] As previously mentioned, the image captured by camera 201 can be an HDR image. Alternatively, the camera can capture several images at different exposure levels. In this case, receiving unit 401 can be further configured to combine the several images to form a single image exhibiting high dynamic range.

[0037] Furthermore, the receiving unit 401 can be further configured to receive multiple images, respectively from each of the optional cameras 202-204. The receiving unit 401 can be further configured to perform the same optional processing as that performed on the images from camera 201. Additionally, the receiving unit 401 can be further configured to combine all images received from cameras 201-204. For example, the receiving unit 401 can be configured to stitch all images from cameras 201-204 together into a single image for input to the determining unit 402. In this way, the determining unit 402 only needs to perform damage determination on a single image. Alternatively, the receiving unit 401 can, instead of outputting, the images received from cameras 201-204 separately to the determining unit 402. In this way, damage determination is performed on single images received from different cameras.

[0038] Optionally, when multiple images are stitched into a single image, the receiving unit 401 can be further configured to synchronize the images, such that when vehicle features appear in multiple images (e.g., the front bumper appears in two images), the images are appropriately combined so that the vehicle features (e.g., the front bumper) are mapped to multiple images. In this way, the generated image includes only vehicle features that appear in the real world, without unnecessary duplication due to using multiple images as sources. For example, simply stitching together two images of the front of vehicle 100 might result in the front bumper appearing twice in the final image. However, by synchronizing the images (e.g., by cropping, rotating, and / or adjusting the perspective of the images), a final image containing only a single front bumper, corresponding to a vehicle in the real world, can be generated.

[0039] Optionally, the receiving unit 401 may further include a camera triggering unit (not shown), configured to trigger camera 201 and / or several cameras 201-204 to capture images of vehicle 100 in response to specific conditions. For example, the camera triggering unit may be configured to capture images only when vehicle 100 has completely stopped in front of the camera, thereby minimizing motion blur of vehicle 100. Optionally, the camera triggering unit is implemented as a statistical method (neural network) configured to infer the presence of a vehicle from camera input. To achieve this, the statistical method is trained on manually labeled images collected from the cameras.

[0040] More specifically, the camera triggering unit can be trained to locate vehicle 100 in the captured images. Furthermore, the camera triggering unit can further determine whether the vehicle is properly positioned, such as for damage detection. To achieve this, a statistical model based on a convolutional neural network can be used—trained from hand-annotated images of properly positioned vehicles and poorly positioned / missing vehicles. Additionally, the camera triggering unit can determine whether the vehicle is stationary: a vehicle is considered stationary if it has not moved between images after several images have been captured. For example, three images of the vehicle can be captured, and if the vehicle has not moved between the first, second, and third images, it can be considered stationary.

[0041] The image processed by the receiving unit 401 is output to the determining unit 402 and the optional training unit 403.

[0042] The determining unit 402 is configured to determine damage based on an image received from the receiving unit 401. This can be done using statistical and / or machine learning models, which are configured to find similarities between portions of the received image and known damage types of the determining unit 402.

[0043] Statistical models can receive a single image or several images to determine damage to vehicles within those images. Optionally, when an image is mistakenly triggered, the statistical model can indicate that a vehicle does not exist in the image, rather than attempting to determine damage to a vehicle that is not present in the image. Statistical models can take the form of standard machine learning models, such as artificial neural networks or Gaussian process classifiers.

[0044] It is conceivable that statistical models can be implemented as deep convolutional neural networks, such as RetinaNet.

[0045] In this way, the determining unit 402 can automatically identify damaged vehicle parts based on images of the vehicle and optionally include the severity of the damage. In this regard, it is envisioned that the determining unit 402 can determine that no damage exists on the vehicle; alternatively, the determining unit 402 can identify at least one damaged area on the vehicle and optionally provide a description of the type of damage (e.g., scratches, dents, or missing parts), the severity of the damage, etc. In this way, the images from the receiving unit 401 are used to provide a detailed list of damages present on the vehicle, to classify the damage and locate it on the vehicle. Furthermore, the determining unit 402 can additionally or alternatively identify which part of the vehicle is damaged, for example, the left front door is damaged.

[0046] Optional training unit 403 is configured to receive images from receiving unit 401. The images are further annotated to indicate image areas in the images corresponding to damage on vehicle 100. Typically, this annotation is performed by an operator to identify image areas indicating damage (whether dents, missing parts, and / or scratches). In this way, training unit 403 receives several images that have been annotated with damage, thus providing some examples of such damage.

[0047] The optional training unit 403 can be further configured to train the determination unit 402 to determine damage based on the received labeled image.

[0048] In this way, training unit 403 can be initially used to train determination unit 402 to identify damage. However, once training is complete, training unit 403 is no longer needed. In this respect, training of determination unit 402 can occur "offline," i.e., before the damage detection system takes effect. In this way, a damage determination system fully trained by training unit 403 can be provided, so that training unit 403 does not need to be activated during operation, as it has already served its purpose of training determination unit 402.

[0049] Therefore, once it is determined that unit 402 has been trained by training unit 403, training unit 403 is no longer needed because training is complete. Instead, training unit 403 can continue to receive labeled images to increase the number of corrupted examples that can be used to train the statistical model. Thus, after a certain period of time, such as weekly, monthly, or yearly, the statistical model can be retrained using all the images collected by training unit 403. In this way, the accuracy of the statistical model can be improved over time.

[0050] The comparison unit 404 is configured to receive an indication of damage on the vehicle determined by the determination unit 402. The comparison unit 404 is configured to compare the received indication of damage with previously known indications of damage on the vehicle, and based on this comparison, determine whether new damage has occurred since the last damage inspection of the vehicle. To achieve this, the comparison unit 404 is used in conjunction with a storage unit 405, which is configured to store information about previously determined damage existing on the vehicle.

[0051] Therefore, the comparison unit 404 is configured to receive the determined damage identified by the determination unit 402 and compare it with information stored in the storage unit 405 regarding previous damage to the vehicle. When the comparison unit 404 does not detect new damage, it can indicate that the determination has been reached. Alternatively, when new damage is determined, the comparison unit 404 can indicate the severity of the new damage and its location on the vehicle. Optionally, the comparison unit 404 can determine whether the new damage requires repair. To achieve this, the comparison unit 404 can be configured to compare the new damage with a certain threshold. When the new damage is determined to be below the certain threshold, it can be classified as "wear," which is common damage caused by operating the vehicle. However, when the damage exceeds the certain threshold, the damage may require immediate correction and repair. Furthermore, the comparison unit 404 can be further configured to objectively inspect each damaged area, even if it has been classified as "wear," and determine that the damage requires immediate correction and repair when the severity of the damage exceeds a certain threshold. In this way, damage continuously classified as "wear" will not steadily increase over time, because only a small amount of damage increases between inspections by the damage detection unit 400. This allows for the assessment of the current severity of the damage and comparison with previously stored damage.

[0052] The comparison unit 404 can be further configured to update the information stored in the storage unit 405 once the comparison is complete. Specifically, the storage unit 405 can be updated by the comparison unit 404 with information about the current state of damage on the vehicle. In this way, the next time the comparison unit 404 compares the damage on the vehicle with the damage stored in the storage unit 405, the comparison is based on the latest information about the state of damage on the vehicle. It is also conceivable that if a workshop or other facility repairs the damage on the vehicle, the workshop can be configured to update the storage unit 405 with information indicating that the damage on the vehicle has been repaired. Accordingly, the next comparison performed by the comparison unit 404 is based on a vehicle without damage.

[0053] Furthermore, storage unit 405 can further store information about the vehicle and associated damage. For example, the vehicle's license plate number / license plate can be stored along with information about damage on the vehicle. It is envisioned that determination unit 402 can be further configured to automatically determine the vehicle's license plate number / license plate and information about identified damage on the vehicle based on an image of vehicle 100. Therefore, storage unit 405 can store previously identified damage and license plate number / license plate, allowing the vehicle to be accurately identified during comparison by comparison unit 404. Alternatively, other identifiers can be used to indicate which specific vehicle is being targeted in the damage detection process. For example, some vehicles have individual numbers or letters painted on their roofs; therefore, it is envisioned that a camera could capture an image of this identifier to automatically identify the vehicle.

[0054] Optionally, a command unit 406 is provided, configured to issue commands to the orientation unit 205. In this regard, the issued command may cause the orientation unit to instruct the vehicle to be driven to the workshop. Alternatively, the issued command may cause the orientation unit to instruct the vehicle to travel on its normal route without proceeding to the workshop, for example, the vehicle will proceed to refuel, restock products, pick up the next customer, etc. As previously mentioned, the orientation unit 205 may take the form of a traffic light / traffic signal / stop light. In this way, a notification can be conveyed to the driver of the vehicle instructing them on how to proceed. Alternatively, for vehicles capable of autonomous driving, the orientation unit 205 may take the form of a transmitter, configured to send a notification to the autonomous vehicle to directly set the vehicle's destination and whether it will proceed to the workshop.

[0055] For this purpose, when the damage comparison performed by the comparison unit 404 indicates at least one of the following: no new damage, existing damage below the repair threshold, and new damage below the repair threshold (i.e. classified as "wear"), the command unit 406 can be configured to issue a notification to bypass the workshop, refuel, refill, pick up the next customer, etc.

[0056] However, when the comparison unit 404 indicates that existing damage exceeds the repair threshold and / or new damage is greater than the repair threshold (i.e., the new damage is serious enough to require immediate repair), the command unit 406 can be configured to issue a notification to proceed to the workshop for repair.

[0057] In this way, the damage detection unit 400 can be configured to identify damage on the vehicle and notify the vehicle to perform repairs if necessary.

[0058] The aforementioned determining unit 402 can be further configured to pay particular attention to damage to specific areas of the vehicle. For example, as... Figure 4 As shown, the determining unit 402 can be trained to focus on damage in areas 501–505. These areas 501–505 can be selected to correspond to areas known to be sensitive to damage. For example, areas 501 and 505 correspond to the bumper areas of vehicle 100, which are most likely to be impacted / scratched in any collision between vehicle 100 and an obstacle. Therefore, they are most likely to be damaged. Areas 501 and 505 also include headlights and turn signals, so damage to these areas may also prevent the proper operation of these devices. In addition, wheel areas 502 and 504 may be rubbed along the curb, resulting in damage to the wheel arches, rims, and tires. Damage to any of these areas may in turn lead to tire leaks and other tire damage, thereby preventing proper handling and driving of the vehicle.

[0059] Other areas of concern may include the door area 503, which is a frequently used area during loading and unloading of vehicle 100. Furthermore, since the doors typically swing outwards from vehicle 100, they are likely to collide with obstacles located on the sides of vehicle 100. Damage to these areas could prevent the doors from opening / closing / locking properly.

[0060] For this purpose, the training of the determination unit 402 can be configured to provide more training on damage occurring in specific regions 501-505, thus making the determination unit 402 more sensitive to damage in these regions 501-505. Additionally or alternatively, the determination unit 402 can be configured to determine damage in specific regions 501-505. For example, the determination unit 402 can be further configured to identify specific regions 501-505 on an image of the vehicle and then determine damage in these regions, without determining damage to any other parts of the vehicle 100. In other words, the determination unit 402 can be configured to detect specific features of the vehicle 100 (e.g., the left door, the front bumper, etc.). Furthermore, the determination unit 402 can further determine damage in / on specific features of the vehicle 100. In this way, the areas of the vehicle 100 most likely to be severely damaged have the damage determined therefrom. Additionally or alternatively, once specific regions 501-505 have been searched for damage and no damage was found, the determination unit 402 can be further configured to search for damage in other locations of the vehicle 100.

[0061] Furthermore, by identifying damage in specific areas 501–505, the damage can be limited to specific parts, enabling the generation of detailed reports and specific repairs. For example, focusing on the front bumper of a vehicle and the damage it contains can provide options for repairing each individual point of damage or, more simply, replacing the entire front bumper. In this way, by focusing on the detection of damage to specific vehicle features, such as headlights, bumpers, and doors, simpler repairs can be facilitated.

[0062] Figure 5 The process performed by the damage detection system according to a first embodiment of the present invention is illustrated. Specifically, flowchart S500 shows the calculation of damage on the vehicle based on an image of the vehicle and information about damage previously detected on the vehicle.

[0063] In step S501, the receiving unit receives an image of the vehicle from the camera. In this respect, only one camera is needed to detect damage on the vehicle. However, it may be advantageous to use multiple images of the vehicle taken from different angles and under different lighting conditions to fully detect all existing damage. Furthermore, HDR images can be used to further improve the contrast of damaged areas on the vehicle. The receiving step S501 can further process the received image to account for different lighting conditions and objects in the image frames other than the vehicle, such as the ground, sky, or people possibly behind the vehicle. For this purpose, the receiving step S501 can crop out such irrelevant objects from the vehicle image so that the further steps in flowchart S500 are not confused by irrelevant objects and / or provide a uniform image that does not vary between vehicles except for the vehicle itself.

[0064] In step S502, damage on the vehicle is determined based on the image received from step S501. To achieve this, a machine learning model and / or a statistical model can be used to identify features of the damage in the image of the vehicle. More specifically, the machine learning model and / or statistical model can be trained on labeled images that have already been manually annotated to identify damaged areas. Based on the labeled images, the model can be used to predict damaged areas on the vehicle based on experience learned from the labeled images, without further manual annotation of any images. The model can be trained before being used as a damage detection system (i.e., training can occur during manufacturing), making an initial training step unnecessary. Furthermore, more labeled images can be collected over time and used to further improve the accuracy of detecting damage in the received images.

[0065] Step S502 can be further configured to determine a vehicle identifier from the received image. For example, the image may include a license plate, which is unique to each vehicle. In this way, detected damage can be associated with a specific vehicle.

[0066] In step S503, the retrieval step stores information from the storage unit regarding damage previously identified on the vehicle. In this regard, damage on the vehicle may have been previously identified using the damage detection system. Therefore, retrieval step S503 retrieves the identified damage. In one example, the identified damage is determined by the vehicle's identifier, allowing for recall of damage on a specific vehicle through a lookup based on the vehicle identifier. Furthermore, when damage on the vehicle is repaired or when a new vehicle is put into use, the storage unit may store information indicating that the vehicle is not damaged (because the damage has been repaired or it is a new vehicle). Therefore, when retrieval step S503 retrieves information from the storage unit, the damage detection system will be used on vehicles where no damage exists.

[0067] In step S504, damage is compared based on the damage determined in determination step S502 and retrieval step S503. More specifically, comparison step S504 receives from determination step S502 the damage determined on the current vehicle at the current time based on the received image. Furthermore, comparison step S504 receives damage previously identified on the current vehicle at a previous time and retrieved by retrieval step S503. Therefore, comparison step S504 compares the damage information to determine if any new damage has occurred between the last damage detection performed on the vehicle and the current instance. When it is determined that no new damage has occurred, comparison step can output "No new damage."

[0068] However, when new damage is identified, comparison step S504 can indicate where the new damage occurred and how severe it is. When the severity of the damage is below a specific threshold, the damage can be classified as "wear" and no further action is required. When the damage is above or equal to the specific threshold, the damage may require immediate repair because it may jeopardize the operation of the vehicle (e.g., headlights not working, tire leaks, etc.), and therefore immediate action can be taken.

[0069] Furthermore, specific areas of a vehicle may have different damage thresholds that necessitate action. For example, a damaged tire might pose a greater threat than a damaged headlight. Therefore, when the severity of damage to a tire is less than that of headlight damage but exceeds a tire-specific threshold, the tire may require immediate repair, while headlight repair (if the headlight damage does not exceed a headlight-specific threshold) can be postponed or not performed. In other words, different thresholds can be applied to different parts of a vehicle to determine when repair is needed and when damage is classified as "wear and tear."

[0070] Flowchart S500 may further include an optional command step S505, which commands the orientation unit to orient the vehicle to the workshop based on the comparison step S504. In other words, based on the result of the comparison performed in comparison step S504, method S500 may further command the orientation unit to orient the vehicle to the workshop for repair. In one example, the orientation unit uses the form of a traffic light with arrows to indicate the direction of vehicle travel. When it is determined that there is no new damage or the damage is determined to be "wear" or the damage is below a certain threshold, command step S505 may command the orientation unit to instruct the vehicle to proceed to the loading / unloading / cleaning location. However, when comparison step S504 determines that the damage is greater than a certain threshold, command step S505 may command the orientation unit to orient the vehicle to the workshop for immediate repair of the damage.

[0071] Modifications and variations

[0072] Various modifications and variations can be made to the above-described specific embodiments without departing from the scope of the present invention.

[0073] While vehicle maintenance has been described above, other preventative factors can be considered to avoid the cause of damage in the first place. For example, when delivery / pickup vehicles travel along a particular route, the route selection may be related to the damage the vehicle may suffer along the way. In this way, when several vehicles travel along the same route and / or portions of the route and cause similar damage at similar locations, the route / part of the route can be marked as hazardous. In this way, route determination can take into account the damage typically suffered along the route and select a different route, marking that route as hazardous and / or advising drivers to exercise special caution along that route. In this way, the cause of the damage can be eliminated.

[0074] In this respect, the delivery / pickup company can record which vehicle has been assigned to which route and the driver to be used for that route. All of this information can be stored in the previously described storage unit 405, making it available for route determination.

[0075] Similarly, when a driver is assigned to drive a specific vehicle, monitoring damage to the vehicle can be linked to that specific driver. In this way, particularly dangerous or inattentive drivers can be identified and retrained to promote better driving. In car rental companies, this information can be used to determine whether the company should ideally avoid renting vehicles to such individuals. In this way, foreseeable vehicle damage from drivers, etc., can be avoided.

[0076] Regarding specific implementations of the computer implementation, the provided specification may describe the steps by which one would modify the computer to implement the system or method. The specific problem to be solved may be in the context of computer-related issues, and the system may not necessarily be achievable solely by manual means or as a series of manual steps. In the context of some specific implementations, computer-related implementations and / or schemes may be advantageous; at least for reasons of providing scalability (managing large amounts of input and / or activity using a single platform / system); the ability to quickly and efficiently aggregate information from different networks; improved decision support and / or analysis, which would otherwise be infeasible; the ability to integrate with external systems whose interface is the only point of connection to the computer implementation; the ability to achieve cost savings through automation; the ability to dynamically respond to and consider updates in various contexts (e.g., rapidly changing instruction or material flow conditions); the ability to apply complex logical rules that are not feasible by manual means; the ability for instructions to be truly anonymous; and so on.

[0077] Using electronic and / or computerized means can provide a platform that is more convenient, scalable, efficient, accurate, and / or reliable than traditional non-computerized means. Furthermore, the system can be computerized and the platform can be advantageously designed for interoperability, where manual operation may be difficult and / or impossible. Moreover, even if feasible, manual operation is unlikely to achieve comparable efficiency and / or reliability.

[0078] Scalability can be useful because it can be advantageous to provide a system that can efficiently manage large amounts of input, output, and / or interconnection and / or integration with external systems.

[0079] In the context of instruction fulfillment, the convenience and effectiveness of a scheme can be very useful, as individuals may have more information to make better instruction and / or fulfillment decisions.

[0080] This system and method can be practiced in various specific embodiments. Appropriately configured computer equipment and associated communication networks, devices, software, and firmware can provide a platform for implementing one or more specific embodiments as described above. By way of example, Figure 6 A computer device 600 is shown, which may include a central processing unit (“CPU”) 602 connected to a storage unit 604 and random access memory 606. The CPU 602 may process an operating system 601, application programs 603, and data 623. The operating system 601, application programs 603, and data 623 may be stored in the storage unit 604 and loaded into the memory 606 as needed. The computer device 600 may further include a graphics processing unit (GPU) 622 operatively connected to the CPU 602 and the memory 606 to offload intensive image processing computations from the CPU 602 and to run these computations in parallel with the CPU 602. An operator 607 may interact with the computer device 600 using a video display 608 connected by a video interface 605, and various input / output devices, such as a keyboard 615, a mouse 612, and a disk drive or solid-state drive 614, connected by an I / O interface 609. In a known manner, mouse 612 can be configured to control the movement of a cursor on video display 608 and to operate various graphical user interface (GUI) controls appearing on video display 608 using mouse buttons. Disk drive or solid-state drive 614 can be configured to accept computer-readable media 616. Computer device 600 can form part of a network via network interface 611, which allows computer device 600 to communicate with other appropriately configured data processing systems (non-display). One or more different types of sensors 635 can be used to receive input from various sources.

[0081] This system and method can be implemented on any type of computer device, including desktop computers, laptop computers, tablet computers, or wireless handheld devices. The system and method can also be implemented as a computer-readable / usable medium including computer program code to enable one or more computer devices to implement each of the processing steps in the method according to the invention. In the case of multiple computer devices performing all operations, the computer devices are networked to distribute the individual steps of the operation. It should be understood that the terms computer-readable medium or computer-usable medium include one or more physical embodiments of program code of any type. Specifically, a computer-readable / usable medium may include program code embodied on one or more portable storage articles (e.g., optical discs, magnetic disks, magnetic tapes, etc.) and embodied on data storage portions of one or more computing devices, such as memory associated with a computer and / or storage system.

[0082] The mobile application of this invention can be implemented as a web service, wherein the mobile device includes a link for accessing the web service, rather than a native application. The described functionality can be implemented on any mobile platform, including Android, iOS, Linux, or Windows.

[0083] In other respects, this disclosure provides systems, apparatus, methods, and computer programming products for implementing the method and enabling the aforementioned functions, including non-transitory machine-readable instruction sets.

[0084] The foregoing description of specific embodiments of the invention has been presented for illustrative and descriptive purposes. It is not intended to be exhaustive or limiting of the invention to the precise forms disclosed. Modifications and variations may be made without departing from the spirit and scope of the invention.

Claims

1. A damage detection unit configured to detect damage on an autonomous vehicle (100), the damage detection unit comprising: A receiving unit (401) is configured to receive images from a camera (201); The determining unit (402) is configured to use at least one of a machine learning model or a statistical model to determine damage on the vehicle (100) based on the received image, the machine learning model or statistical model having been trained on an annotated image that has been annotated to indicate the damaged area on the vehicle (100); Storage unit (405) is configured to store information about damage previously determined to exist on the vehicle (100); A comparison unit (404) is configured to compare the damage determined by the determining unit (402) with information about previous damage stored in the storage unit (405); as well as A command unit (406) is configured to issue one or more commands to a direction unit (205), the direction unit (205) including a transmitter configured to send a notification to the autonomous vehicle (100) when the determined damage is greater than a predetermined threshold, to directly set the destination of the vehicle (100) to travel to the workshop so that the damage can be repaired.

2. The damage detection unit according to claim 1, characterized in that, The comparison unit (404) is further configured to determine whether new damage has occurred based on damage compared with information about previous damage.

3. The damage detection unit according to claim 1 or 2, wherein the damage detection unit further comprises a training unit (403) configured to receive an annotated image from the receiving unit (401) to indicate a damaged area on the vehicle, and configured to train the determining unit (402) to determine damage based on the annotated image.

4. The damage detection unit according to claim 1 or 2, wherein the damage detection unit further comprises a camera triggering unit, which includes a machine learning model configured to cause the camera (201) to capture an image of the vehicle (100).

5. A damage detection system, the system comprising: A camera (201) configured to capture images of a vehicle; and Damage detection unit according to any of the prior claims.

6. The damage detection system according to claim 5, wherein the system further includes the orientation unit (205).

7. A damage detection method for detecting damage on an autonomous vehicle, the method comprising the following steps: Receive images from camera (201); Damage on the vehicle (100) is determined based on the received images and using at least one of a machine learning model or a statistical model, which is trained on labeled images that have been labeled to indicate areas of damage on the vehicle (100); Retrieve information regarding previously determined damage to the vehicle (100); and The damage determined in the determination step is compared with the information about previous damage retrieved in the retrieval step; One or more commands are issued to a direction unit (205), the direction unit (205) including a transmitter configured to transmit instructions to the autonomous vehicle (100) when a determined degree of damage is greater than a predetermined threshold, to directly set the destination of the vehicle (100) to travel to the workshop so that the damage can be repaired.

8. The damage detection method according to claim 7, characterized in that, The method further includes determining whether new damage has occurred based on comparison-based damage assessment.

9. The damage detection method according to claim 7 or 8, wherein the method further comprises: Receive labeled images to indicate damaged areas on the vehicle (100); and Train at least one of a machine learning model or a statistical model to determine damage based on the labeled image.

Images