Edge systems, vehicles and methods

Abstract

An edge system for a vehicle is provided. The edge system comprises a drive path image capture unit configured to capture one or more images of a drive path of the vehicle. The edge system comprises a site-of-interest (SOI) camera configured to capture one or more images of an SOI monitored by the vehicle. The edge system comprises processing circuitry configured to apply a first artificial intelligence (AI) object recognition model to the one or more images captured by the drive path image capture unit to recognise one or more objects in the drive path of the vehicle. The first AI recognition model is adapted for recognising the objects in the drive path of the vehicle. The processing circuitry is configured to apply a second AI object recognition model to the one or more images captured by the SOI image capture unit to recognise one or more objects in the SOI. The second AI object recognition model is adapted for recognising the objects in the SOI. The processing circuitry is configured to provide an indication of the recognised objects in the drive path of the vehicle to a drive control unit for the vehicle. The processing circuitry is configured to provide an indication of the recognised objects in the SOI to a wireless transmission unit comprised in the edge system. The wireless transmission unit is configured to transmit the indication of the recognised objects in the SOI via a wireless communications network.

Description

[0001] EDGE SYSTEMS, VEHICLES AND METHODS

[0002] BACKGROUND

[0003] Field of Disclosure

[0004] The present disclosure relates to edge systems, vehicles and methods.

[0005] The present application claims Paris Convention priority from European patent application number 24160376.0, filed on 28 February 2024, the contents of which are hereby incorporated by reference in their entirety.

[0006] Description of Related Art

[0007] The “background” description provided herein is for the purpose of generally presenting the context of the disclosure. Work of the presently named inventors, to the extent it is described in this background section, as well as aspects of the description which may not otherwise qualify as prior art at the time of filing, are neither expressly or impliedly admitted as prior art against the present invention.

[0008] Persons involved in a disaster (for example, an accident such as car crash, boat crash or natural disaster such as an earthquake) may be in need of urgent medical assistance or rescue. Typically, either a person involved in the disaster, or a witness, will contact rescue services (such as the fire brigade or coastguard) to inform them of the type of disaster and location. The rescue services will then urgently proceed to the location of the accident, determine the extent of the disaster, locate injured persons and treat the injured persons and / or prepare the injured persons for transport. The rescue services may also locate active sources of danger (such as a fire on a burning car) and take action to make the area safe (e.g. by extinguishing the fire). However, the information provided to the rescue services is limited by what the witness has seen. The witness may not accurately assess the situation and may leave out important information. For example, the witness may not have found everyone who was injured, the witness may incorrectly assess the severity of the situations or leave out important details such as the location of active sources of danger. This is particularly prevalent in situations where visibility is poor, such as in a forest. Therefore, the ability of the rescue services to quickly locate objects (such as persons or sources of danger) in a disaster zone is limited by the information provided by the witness.

[0009] Infrared cameras have been deployed in helicopters to locate objects in disaster zones and report the locations of those objects to the rescue services. However, a helicopter with infrared cameras is extremely expensive and heavy.

[0010] Additionally, smartphone tracking has been used to track the locations of persons in a disaster zone. However, this requires that the users have a smartphone near them which has battery power and does not provide an overview of the disaster zone.

[0011] Unmanned aerial vehicles (UAVs) have also been deployed to assist rescue workers during medical emergencies. For example, drones have been equipped with medical supplies and steered towards injured persons in difficult-to-reach places. Furthermore, drones have been deployed with cameras so that a remote operator can monitor a site of interest (SOI) traversed by the drone. For example, police drones may be used to monitor for missing powers, weapons, or other suspicious objects by flying over towns, capturing images and transmitting the images back to a control center for manual observation. However, such images may contain sensitive information such as GDPR sensitive data and it may be difficult to use the images to locate objects (such as persons) in the images because the objects may have moved since the images were captured.

[0012] In addition to the above-mentioned disadvantages, existing techniques for disaster zone observation include the following disadvantages: it takes a long time to overview the total area of a disaster zone and, due to limited site overview, injured persons may be overseen.

[0013] There is therefore a desire for improved systems, vehicles and methods which can aid rescue workers in quickly finding objects involved in a disaster.

[0014] SUMMARY OF THE DISCLOSURE

[0015] The present disclosure can help address or mitigate at least some of the issues discussed above.

[0016] Respective aspects and features of the present disclosure are defined in the appended claims.

[0017] It is to be understood that both the foregoing general description and the following detailed description are exemplary, but are not restrictive, of the present technology. The described embodiments, together with further advantages, will be best understood by reference to the following detailed description taken in conjunction with the accompanying drawings.

[0018] BRIEF DESCRIPTION OF THE DRAWINGS

[0019] A more complete appreciation of the disclosure and many of the attendant advantages thereof will be readily obtained as the same becomes better understood by reference to the following detailed description when considered in connection with the accompanying drawings wherein like reference numerals designate identical or corresponding parts throughout the several views, and wherein:

[0020] Figure 1 schematically illustrates an example of an edge system for a vehicle in accordance with example embodiments;

[0021] Figure 2 schematically illustrates an example of an edge system for a vehicle in accordance with example embodiments;

[0022] Figure 3 schematically illustrates an example of a vehicle comprising an edge system in accordance with example embodiments;

[0023] Figure 4 schematically illustrates a flight path of a UAV in accordance with example embodiments;

[0024] Figure 5 schematically illustrates an example of an edge system in accordance with example embodiments;

[0025] Figure 6 schematically illustrates an example of an edge system in accordance with example embodiments;

[0026] Figure 7 is a flow diagram illustrating a method of operating an edge system for a vehicle in accordance with example embodiments;

[0027] Figure 8 is a flow diagram illustrating a method of operating an edge system in accordance with example embodiments.

[0028] DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] The foregoing paragraphs have been provided by way of general introduction, and are not intended to limit the scope of the following claims. The described embodiments, together with further advantages, will be best understood by reference to the following detailed description taken in conjunction with the accompanying drawings (wherein like reference numerals designate identical or corresponding parts throughout the several views).

[0030] Al-based Object Recognition

[0031] As will be understood by a person skilled in the art in the field of computer vision, object recognition describes the computational processes used to identify objects in digital images. Object recognition may refer to image classification, object localisation or object detection. Image classification refers to sorting images into classes. Object localisation refers to locating objects within an image and demarcating the object in the image with, for example, a bounding box. Object detection refers to locating objects within an image, demarcating the object in the image with, for example, a bounding box, and sorting the objects in the image into classes.

[0032] Object recognition is recognised as an important problem in the field of computer vision and has a wide variety of applications in areas such as surveillance and security, autonomous driving, or other applications which employ object tracking or segmentation.

[0033] In recent years, artificial intelligence algorithms have been applied with great success to object recognition tasks. As Al continues to develop and improve, object recognition is becoming increasingly effective. One branch of Al which has had particular success in object recognition is “machine learning”. As will be understood by a person skilled in the art, machine learning models may use supervised learning, unsupervised learning and / or reinforcement learning.

[0034] An example of a supervised learning model will now be described. A supervised learning model is trained using labelled training data to learn a function that maps inputs (typically provided as feature vectors) to outputs (i.e. labels). The labelled training data comprises pairs of inputs and corresponding output labels. The output labels are typically provided by an operator to indicate the desired output for each input. The supervised learning model processes the training data to produce a function that can be used to map new (i.e. unseen) inputs to a label. The input data (during training and / or inference) may comprise various types of data, such as numerical values, images, video, text, or audio. Raw input data may be pre-processed to obtain an appropriate feature vector used as input to the model - for example, features of an image (such as edges) may be extracted to obtain a corresponding feature vector. It will be appreciated that the type of input data and techniques for preprocessing of the data (if required) may be selected based on the specific task the supervised learning model is used for. Once prepared, the labelled training data set is used to train the supervised learning model. During training the model adjusts its internal parameters (e.g. weights) so as to optimize (e.g. minimize) an error function, aiming to minimize the discrepancy between the model’s predicted outputs and the labels provided as part of the training data. In some cases, the error function may include a regularization penalty to reduce overfitting of the model to the training data set. The supervised learning model may use one or more machine learning algorithms in order to learn a function which provides a mapping between its inputs and outputs. Example suitable learning algorithms include linear regression, logistic regression, artificial neural networks, decision trees, support vector machines (SVM), random forests, and the K-nearest neighbour algorithm. Once trained, the supervised learning model may be used for inference - i.e. for predicting outputs for previously unseen input data. The supervised learning model may perform classification and / or regression tasks. In a classification task, the supervised learning model predicts discrete class labels for input data, and / or assigns the input data into predetermined categories. In a regression task, the supervised learning model predicts labels that are continuous values.

[0035] Unsupervised learning models differ from supervised learning models in that the training data is not labelled. Unsupervised models are therefore suited to discovering new patterns and relations in raw unlabelled data whereas supervised learning is more suited to learning relationships between input data and output labels. An autoencoder is an example of an unsupervised machine learning model.

[0036] In reinforcement learning, an agent interacts with an environment by performing actions and learns from the results of its actions based on feedback, thereby enabling the agent to progressively improve its decision making. The reinforcement learning algorithm may rely on a model of the environment (e.g. based on Markov Decision Processes (MDPs)) or be model-free. Example suitable model-free reinforcement learning algorithms include Q- learning, SARSA (State-Action-Reward-State-Action), Deep Q-Networks (DQNs), or Deep Deterministic Policy Gradient (DDPG).

[0037] Deep Learning

[0038] As will be understood by a person skilled in the art, “deep learning” is a specialised form of “machine learning”. Deep learning algorithms have had particular success in performing object recognition tasks. In particular, deep learning algorithms have been developed which are considered to be robust to occlusion, and able to cope with complicated scenes and difficult illumination.

[0039] Deep learning models are based on artificial neural networks with representation learning. Representation learning allows a model to discover representations needed for feature detection or classification from raw data. This is in contrast to other machine learning models where, for example, features are manually extracted from images and supplied as input data to the model. Accordingly, deep learning models can use images as the input and do not require manual extraction of features from the image. An example of using a supervised deep learning model to perform object recognition is discussed below:

[0040] In order to perform an object recognition task, a supervised deep learning model may be trained based on training data such as a set of images labelled with known objects. For example, input data (i.e. an image containing an object) and output labels (i.e. an identity and location of the object in the image) are provided to the deep learning model. The deep learning model processes the training data to learn a function which maps the input data to the output labels. Then, when new input data (such as a new image containing an unknown object) is provided to the deep learning model, the deep learning model uses the function to provide output labels (such as a location and identity of the unknown object in the new image) based on the new input data. Although the above example discusses a supervised deep learning model, it will be appreciated by a person skilled in the art that deep learning models may additionally, or alternatively, use unsupervised learning and / or reinforcement learning.

[0041] As will be known to one skilled in the art, particularly successful deep learning models in the field of object recognition include the Region-Based Convolutional Neural Network (R-CNN) family of models, and the You Only Look Once (YOLO) family of models. The R-CNN family of models includes the R-CNN model, the Fast R-CNN model and the Faster R-CNN model. The YOLO family of models contains eight different YOLO versions, the most recent version being YOLO v.8. Although the R-CNN family of models may generally be more accurate than the YOLO family of models, the YOLO family of models are much faster and therefore allow object recognition in real time. In preferred embodiments, the Al models used for object recognition use the YOLOv8 model. Although the use of deep learning models have many advantages as discussed above, the use of deep learning models is computationally intensive. However, recent advances in computer processor technology has meant that deep learning can be employed on relatively low power or low complexity devices or systems such as edge Al cameras.

[0042] As will be appreciated by a person skilled in the art, edge Al camera can provide image analysis and / or video analytics at the edge of a communications network (thus maximizing network and bandwidth efficiencies). In other words, the processing (such as Artificial Intelligence (Al) processing, utilizing the use of a trained model) can be performed at the edge of a communications network.

[0043] The first and second Al object recognition models referred to in the subsequent description may be implemented using any of the aforementioned Al object recognition models.

[0044] Figure 1 is an example of an edge system 200 for a vehicle 300 in accordance with example embodiments. The vehicle 300 may be an unmanned terrestrial vehicle (UAV). A UAV is a vehicle which is operable to fly and which does not require human presence on-board for flight control. Examples of UAVs include drones, or satellites. A UAV may be non- autonomous. For example, a drone which is entirely remote-controlled. Alternatively, a UAV may be semi-autonomous, where the flight control is partially implemented by instructions transmitted to, or configured in, the drone and partially implemented of the drones’ own volition according to decisions made by the drone.

[0045] The edge system 200 comprises a drive path image capture unit 210 configured to capture one or more images of a drive path of the vehicle. For example, the images of the drive path of the vehicle may be images in the drive direction of the vehicle.

[0046] The edge system 200 comprises a site-of-interest (SOI) image capture unit 220 configured to capture one or more images of an SOI monitored by the vehicle. The SOI may be for example, a disaster zone such as a vehicle crash zone, a rescue zone or the like, which the vehicle 300 is monitoring. In an example, where the vehicle 300 is a UAV, the drive path camera may be on the front of the UAV to capture images in the flight direction of the UAV whereas the SOI image capture unit 220 may be located on the underside of the UAV to capture images of a disaster zone. There may be a plurality of SOI image capture units in an edge system which widens the field of view and therefore allows a wider disaster zone to be captured. Accordingly, it may take a shorter time to capture images over the total area of the disaster zone. Furthermore, the risk of missing important objects (such as injured persons) is reduced. A wider field of view enables a fast overview of a disaster zone to be provided.

[0047] Although not shown in Figure 1 , the edge system may comprise one or more additional SOI image capture units each configured to capture one or more images of the SOI.

[0048] Although not shown in Figure 1 , each of the SOI image capture unit 220 and drive path image capture unit 210 may each comprise a focusing element for focusing light from the scene and a detector for converting the light from the scene into image data. The focusing element may be a reflective or refractive element. For example, the focusing element may comprise a lens, or lens system, for focusing the light from the scene onto the detector. The detector may, in examples, be a type of image sensor such as a charge-coupled device (CCD) or a complementary metal-oxide-semiconductor (CMOS). The detector may provide image data in any suitable format depending on the situation to which the embodiments of the present disclosure are applied. In examples, the SOI image capture unit 220 and drive path image capture unit 210 may acquire high definition image data (such 4K, 8K or higher resolutions).

[0049] The edge system 200 comprises a system processing unit 240. The system processing unit 240 is configured to apply a first Al object recognition model to the one or more images captured by the drive path image capture unit 210 to recognise one or more objects in the drive path of the vehicle 300. The objects recognised in the images captured by the drive path image capture unit 210 may comprise obstacles in the drive path such as such as trees, pedestrians, buildings or the like. The system processing unit 240 may be configured to retrieve the images captured by the drive path image capture unit 210 from the drive path image capture unit 210, or from the system memory 230.

[0050] The first Al recognition model is adapted for recognising the objects in the drive path of the vehicle. For example, in embodiments where the vehicle is a UAV, the first Al object recognition model may be adapted for a flight path object recognition task. To achieve this, the first Al object recognition model may be trained based on flight path images such as images taken from the perspective of a UAV. For example, the images may include clouds, tall trees, tall buildings as seen from a camera at the height of a typical UAV. In some embodiments, where the vehicle 300 is a terrestrial vehicle, the first Al object recognition model may be adapted for a terrestrial path object recognition task. To achieve this, the first Al object recognition model may be trained based on terrestrial path images such as images taken from the perspective of a terrestrial vehicle. For example, the images may include roads, other terrestrial vehicles, pedestrians or the like. The images used to train the first Al object recognition model may be publicly available images to ensure GDPR compliance.

[0051] The system processing unit 240 is configured to apply a second Al object recognition model to the one or more images captured by the SOI camera to recognise one or more objects in the SOI. The system processing unit 240 may be configured to retrieve the images captured by the SOI image capture unit 220, or from the system memory 230. The objects recognised in the images captured by the SOI image capture unit 220 may comprise, for example, injured persons, or persons otherwise in need of rescue, damaged vehicles, or the like. Persons in an image may be recognised based on detection of human body parts for, example. Furthermore, a person in need of urgent rescue may be recognised based on their reaction to the vehicle. Furthermore, the second Al object recognition may detect whether recognised persons are active or inactive.

[0052] The second Al object recognition model is adapted for recognising the objects in the SOI. In other words, the second Al object recognition model is adapted for an SOI object recognition task. To achieve this, the second Al object recognition model may be trained based on images of similar SOIs. For example, if the SOI is a vehicle crash, the second Al object recognition model may be trained based on previous images of vehicle crashes. In another example, if the SOI is a burning building, the second Al object recognition model may be trained based on previous images of burning buildings. The images used to train the second Al object recognition model may be publicly available images to ensure GDPR compliance.

[0053] Therefore, the first Al object recognition model and the second Al object recognition model may be adapted for performing object recognition in images of different types of scenery. The first and second Al object recognition models may be the same type of model (e.g. the first and second Al object recognition models may be deep learning models such as YOLO v8) or different types of models.

[0054] The system processing unit 240 may be a microprocessor carrying out computer instructions or may be an Application Specific Integrated Circuit. Computer instructions may be stored on storage medium (such as the system memory 230). The computer instructions may be embodied as computer software that contains computer readable code which, when loaded onto the system processing unit 240, configures the system processing unit 240 to perform a method according to embodiments of the disclosure.

[0055] In embodiments where the edge system 200 comprises one or more additional SOI image capture units, the system processing unit 240 may be configured to apply one or more additional Al object recognition models to the images captured by the one or more additional SOI image capture units respectively. In other words, a different Al object recognition model is applied to the images captured by different SOI image capture units. The second Al object recognition model and the one or more additional Al object recognition models form a plurality of Al object recognition models each adapted for recognising objects in different types of SOI. For example, one of the plurality of Al object recognition models may be adapted for object recognition in SOIs with vehicle crashes because it was trained on images of vehicle crashes and another of the plurality of object recognition models may be adapted for object recognition in SOIs with burning buildings because it was trained in images of burning buildings. Therefore, the type of SOI may be referring to the type of incident occurring in the SOI, the type of terrain (i.e. whether it is an urban area or in the countryside) and the like.

[0056] In some embodiments, the first Al object recognition model and the plurality of Al object recognition models may be YOLOv8 models.

[0057] The edge system 200 may comprise a system memory 230 configured to store the images captured by the drive path image capture unit 210 and the SOI image capture unit 220. The SOI image capture unit 220 and the drive path image capture unit 210 may provide captured images to the system memory 230 for storage. The system processing unit 240 may be configured to control the system memory 230 to delete images captured by the drive path image capture unit 210 after the first Al object recognition model has been applied to the images captured by the drive path image capture unit 210 and to delete the images captured by the SOI image capture unit 220 after the second Al object recognition model has been applied to the images captured by the SOI image capture unit 220. If there are a plurality of SOI image capture units, then the system memory 230 may be configured to store the images captured by the plurality of SOI image capture units and the system processing unit 240 may be configured to control the system memory 230 to delete those images after the plurality of Al recognition models have been applied to those images. By deleting images after applying object recognition, storage requirements can be reduced. Furthermore, GDPR compliance is improved because potentially sensitive data in the images is not retained. In some embodiments, the system memory 230 stores the first Al object recognition model and the second Al object recognition model. In some embodiments, the system memory 230 may also store one or more additional Al object recognition models, where the second Al object recognition model and the one or more additional Al object recognition models form a plurality of Al object recognition models for recognising objects in different types of SOI. In such embodiments, the system processing unit 240 may be configured to select the second Al object recognition model from the plurality of Al object recognition models based on the type of the SOI to be monitored by the vehicle 300. In some embodiments, an operator of the edge system 200 may configure the system processing unit 240 with the second Al object recognition model before deploying the vehicle 300 based on the operator’s knowledge of the type of SOI to be monitored. Therefore, different Al models can be applied to an SOI depending on the type of SOI so that the object recognition may be faster and more accurate.

[0058] The system memory 230 may comprise, for example, a magnetically readable medium, optically readable medium or solid state type circuitry

[0059] The system processing unit 240 is configured to provide an indication of the recognised objects in the drive path of the vehicle 300 to a drive control unit 320 for the vehicle 300. For example, there may be a wired or wireless connection between the system processing unit 240 and the drive control unit 320 for providing the indication. In cases where there is a wireless connection (for example, a Bluetooth connection) between the system processing unit 240 and the drive control unit 320, the system processing unit 240 may forward the indication to the wireless transmission unit 260 (or to another wireless transmission unit in the edge system 200 not shown) for transmission of the indication over the wireless connection.

[0060] The system processing unit 240 is configured to provide an indication of the recognised objects in the SOI to a wireless transmission unit 260 comprised in the edge system 200. For example, the system processing unit 40 may transmit the indication to the wireless transmission unit 260 over a wired connection.

[0061] The wireless transmission unit 260 is configured to transmit the indication of the recognised objects in the SOI via a wireless communications network 500. For example, the indication may be transmitted to a radio access network of the wireless communications network 500. The indication may be transmitted via the wireless communications network 500 to, for example, communications devices belonging to rescue workers or to a control center which co-ordinates the rescue workers. The indication may be transmitted, for example, via a radio access network to a central unit (CU) of a wireless communications network. By transmitting an indication of the recognised objects in the SOI via a wireless communications network 500, rescue workers can be quickly provided with information pertinent to the disaster and co-ordinate their efforts to locate and treat those people in the disaster zone who are in need of the most urgent assistance. In embodiments where there are a plurality of SOI image capture units comprised in the edge system 200, the wireless transmission unit 260 may transit an indication of the objects recognised in the images captured by each of the plurality of SOI image capture units.

[0062] The wireless transmission unit 260 may be implemented by any wireless transmitter or transmitter circuitry, or a transceiver or transceiver circuitry. The drive control unit 320 is a unit configured to control the drive path of the vehicle 300. In some embodiments, as shown in Figure 1 , the drive control unit 320 is comprised in the vehicle 300. In cases where the vehicle 300 is at least semi-autonomous, the drive control unit 320 may adjust the drive path of the vehicle based on the indication of the recognised objects in the drive path of the vehicle 300 provided by the object recognition circuitry. For example, the drive control unit 320 may adjust the drive path of the vehicle 300 to avoid colliding with the recognised objects in the drive path of the vehicle. In some embodiments, the drive control unit 320 may be comprised in a remote control unit for the vehicle 300 (not shown). The remote control unit may be operated by an operator such as a rescue worker for example. In such embodiments, the system processing unit 240 may provide the indication of the recognised objects in the drive path of the vehicle 300 to the wireless transmission unit 260. In such embodiments, the wireless transmission unit 260 is configured to transmit the indication of the recognised objects in the drive path of the vehicle 300 to the drive control unit 230 in the remote control unit. In such embodiments, the operator may manually control the drive path of the vehicle based on the indication of the recognised objects in the drive path of the vehicle 300.

[0063] By configuring the edge system 200 with a first and second Al object recognition models each adapted for a different object recognition task, the object recognition results can be attained faster and more accurately. For example, an Al object recognition model trained on images of a burning building, is likely to more quickly and accurately recognise persons in need of rescue from a burning building compared with an Al object recognition model trained on images of vehicle crashes.

[0064] By configuring the edge system 200 with the system processing unit 240, processing involving the application of the Al object recognition models to images is performed on the edge system 200. By processing the images for object recognition locally at the edge of a communications network, there is no requirement to transmit the images away from the edge system 200 for processing, thereby improving communications efficiency.

[0065] By transmitting an indication (for example, metadata) of the recognised objects in the SOI, rather than transmitting entire SOI images with the recognised objects, communications efficiency can be improved because less information is transmitted over the wireless communications network. The metadata may comprise, for example, one or more of a number of recognised objects in the SOI, a number of humans in the SOI, a number of vehicles in the SOI, an indication of a state of the humans or the vehicles in in the SOI, a number of people who are standing up and / or sitting down, a location of recognised people, number plates of recognised vehicles.

[0066] Furthermore, by transmitting an indication (for example, metadata) of the recognised objects in the SOI, rather than transmitting entire SOI images with the recognised objects, means that GDPR compliance can be improved. For example, the indication of the recognised objects in the SOI may not contain any information which could identify persons in the SOI. Therefore, example embodiments can improve GDPR compliance by avoiding the transmission of GDPR sensitive data.

[0067] In some embodiments, the edge system 200 comprises a spotlight projection unit 270 configured to project one or more spotlights onto one or more of the objects recognised in the SOI. For example, the spotlight projection unit 270 may project a spotlight onto one or more persons who were recognised as injured or otherwise in need of rescue. The spotlights can quickly guide rescue workers to the locations of persons in need of rescue. In some embodiments, the one or more spotlights are coloured laser emitting diode (LED) spotlights. In some embodiments, the colour of a spotlight indicates a state of the recognised object. For example, a green spot light may be projected on a recognised person who is uninjured whereas a red spot light may be projected on a recognised person who is injured or otherwise in need of rescue. In some embodiments, the one or more spotlights are laser spot lights (for example, infrared laser spot lights). The use of infrared spot lights is particularly advantageous when visibility is poor such as when there is fog, or during the night.

[0068] Figure 2 schematically illustrates an alternative configuration of an edge system 600 for a vehicle 300 in accordance with example embodiments.

[0069] The SOI image capture unit 220, drive path image capture unit 210, wireless transmission unit 260, spotlight projection unit 270, the wireless communications network 500, the drive control unit 320, and the vehicle 300 described with reference to Figure 1 also appear in Figure 2. Those objects will not be described again for brevity.

[0070] As shown in Figure 2, the edge system 600 comprises a drive path camera 610. The drive path camera comprises the drive path image capture unit 210, a drive camera memory 612 and a drive camera processing unit 614.

[0071] The drive camera processing unit 614 is configured to apply the first Al object recognition model to the images captured by the drive path image capture unit 210 to recognise one or more objects in the drive path of the vehicle 300. The drive camera processing unit 614 may be configured to retrieve the images captured by the drive path image capture unit 210, or from the drive camera memory 612. The drive camera processing unit 614 is configured to provide an indication of the recognised objects in the drive path of the vehicle to the drive control unit 320 for the vehicle 300. For example, there may be a wired or wireless connection between the drive camera processing unit 614 and the drive control unit 320 for providing the indication. In cases where there is a wireless connection (for example, a Bluetooth connection) between the drive camera processing unit 614 and the drive control unit 320, the drive camera processing unit 614 may forward the indication to the wireless transmission unit 260 (or to another wireless transmission unit in the edge system 200 not shown) for transmission of the indication over the wireless connection.

[0072] The drive camera processing unit 614 may be a microprocessor carrying out computer instructions or may be an Application Specific Integrated Circuit. Computer instructions may be stored on storage medium (such as the drive camera memory 612). The computer instructions may be embodied as computer software that contains computer readable code which, when loaded onto the drive camera processing unit 614, configures the drive camera processing unit 614 to perform a method according to embodiments of the disclosure

[0073] The drive camera memory 612 is configured to store the images captured by the drive path image capture unit 210. The drive camera processing unit 614 may be configured to control the drive camera memory 612 to delete the images captured by the drive path image capture unit 210 after the first Al object recognition model has been applied to the images captured by the drive path image capture unit 210. In some embodiments, the drive camera memory 612 is configured to store the first Al object recognition model.

[0074] The edge system 600 comprises an SOI camera 620. The SOI camera 620 comprises the SOI image capture unit 220, an SOI camera memory 622 and an SOI camera processing unit 624.

[0075] The SOI camera processing unit 624 is configured to apply the second object recognition model to the images captured by the SOI image capture unit 220 to recognise one or more objects in the SOI. The SOI camera processing unit 624 may be configured to retrieve the images captured by the SOI image capture unit 220, or from the SOI camera memory 622. The SOI camera processing unit 624 is configured to provide an indication of the recognised objects in the SOI to the wireless transmission unit 260. For example, the SOI camera processing unit 624 may be configured to transmit the indication of the wireless transmission unit 260 via a wired connection.

[0076] The SOI camera processing unit 624 may be a microprocessor carrying out computer instructions or may be an Application Specific Integrated Circuit. Computer instructions may be stored on storage medium (such as the SOI camera memory 622). The computer instructions may be embodied as computer software that contains computer readable code which, when loaded onto the drive camera processing unit 614, configures the SOI camera processing unit 624 to perform a method according to embodiments of the disclosure

[0077] The SOI camera memory 622 is configured to store the images captured by the SOI image capture unit 220. The SOI camera processing unit 624 may be configured to control the SOI camera memory 622 to delete the images captured by the SOI image capture unit 220 after the second Al object recognition model has been applied to the images captured by the SOI image capture unit 220. In some embodiments, the SOI camera memory 622 may be configured to store the second Al object recognition model. In some embodiments, the SOI camera memory 622 may also store one or more additional Al object recognition models, where the second Al object recognition model and the one or more additional Al object recognition models form a plurality of Al object recognition models for recognising objects in different types of SOI. In such embodiments, the SOI camera processing unit 624 may be configured to select the second Al object recognition model from the plurality of Al object recognition models based on the type of the SOI to be monitored by the vehicle 300. In some embodiments, an operator of the edge system 200 may configure the drive camera processing unit 614 with the second Al object recognition model before deploying the vehicle 300 based on the operator’s knowledge of the type of SOI to be monitored. Therefore, different Al models can be applied to an SOI depending on the type of SOI so that the object recognition may be faster and more accurate.

[0078] Therefore, whereas in Figure 1 , the edge system 200 comprises a system memory 230 and system processing unit 240 for the SOI image capture unit 220 and the drive path image capture unit 210, the edge system 300 in Figure 2 comprises a dedicated SOI camera memory 622 and dedicated SOI camera processing unit 624 in the SOI camera 620 and a dedicated drive camera memory and dedicated drive camera processing unit 614 in the drive path camera 610.

[0079] Each of the memories 230, 612, 622 may be referred to as “memory”.

[0080] Each of the processing units 240, 614, 624 may be referred to as “processing circuitry”.

[0081] In some embodiments, the edge system 200 in Figure 1 or the edge system 600 shown in Figure 2 is comprised in the vehicle 300. In such embodiments, the drive control unit 320 may be comprised in the edge system 200, 600. In some embodiments, the edge system 200 in Figure 1 or the edge system 600 shown in Figure 2 is mounted on or attached to the vehicle 300.

[0082] Figure 3 illustrates an example of a vehicle 400 comprising an edge system in accordance with example embodiments. As shown in Figure 3, an SOI camera 460 captures an image 410 of an SOI. In The example illustrated in Figure 3, the SOI is the scene of a traffic accident. The edge system applies an Al object recognition algorithm to the image 410. In this case, the Al object recognition algorithm has been trained on images of previous traffic accidents and is therefore adapted for performing a traffic accident object recognition task. The Al object recognition recognises three objects in the image 410, namely, a car 420, a person 430 and a motorcycle 440. In the example shown in Figure 3, the edge system comprises a spotlight projection unit. The spotlight protection unit projects a spotlight 450 onto the car 420 as shown in Figure 3. This guides rescue workers (such as firemen and paramedics) to the car 420. The edge system 400 transmits an indication of the recognised objects in the image 440 via a wireless communications network. For example, the edge system 400 may transmit an indication that one car, one person and one motorcycle were detected. The indication may include an indication of whether the person is injured. The indication may include an estimated location of the car, person and motorcycle. However, the indication may not contain information which could identify the person, or the owners of the car and motorcycle, to improve compliance with GDPR and simultaneously reduce the amount of information having to be transmitted. The indication may be transmitted to communications devices of rescue workers or to a control center which is in contact with the rescue workers.

[0083] An example of a flight path for a semi-autonomous drone is schematically illustrated in Figure 4. In Figure 4, the drone is released at a release point 2. The release point 2 may, for example, be a center point of a disaster zone. After release, the drone then follows a first flight path 4 until it reaches a first detection point 6. At the first detection point 6, the drone stops (for example, hovers mid-air) and may transmit data to a control center and project light guiding signals to guide rescue workers to particular locations within the disaster zone. Then the drone follows a second flight path 8 until it reaches a second detection point 10. During the second flight path 8, the drone encounters a first obstacle 12 and a second obstacle 14. In order to avoid a collision with the first obstacle 12 or second obstacle 14, the drone adjusts its flight path to avoid the obstacles. At the second detection point 10, the drone stops (for example, hovers mid-air) and may transmit data to a control center and project light guiding signals to guide rescue workers to particular locations within the disaster zone. Although only two detection points are shown in Figure 3, there may be many more detection points.

[0084] Wearable Edge System

[0085] Figure 5 schematically illustrates an edge system 700 in accordance with example embodiments. The edge system 700 may be a wearable edge system. For example, the edge system 700 may be worn by a rescue worker. Alternatively, the edge system 700 may be comprised in or mounted on a vehicle.

[0086] The edge system 700 comprises a plurality of site-of-interest (SOI) image capture units (comprising a first SOI image capture unit 720 and a second SOI image capture unit 710) each configured to capture one or more images of an SOI. The plurality of SOI capture units may each be broadly similar to the SOI image capture unit 220 described with reference to Figures 1 and 2.

[0087] The edge system 700 comprises a system processing unit 740 configured to apply a first Al object recognition model to the one or more images captured by the first SOI image capture unit 720 to recognise one or more objects in the SOI. The system processing unit 740 is configured to apply a second Al object recognition model to the one or more images captured by the second SOI camera 710 to recognise one or more objects in the SOI. The system processing unit 740 may be configured to retrieve the captured images from the first SOI image capture unit 720 and the second SOI image capture unit 710, or from the system memory 730. The first and second Al object recognitions model are adapted for recognising the objects in different types of SOI. The first and second Al object recognition models may be the same type of model (e.g. the first and second Al object recognition models may be deep learning models such as YOLO v8) or different types of models.

[0088] The system processing unit 740 may be a microprocessor carrying out computer instructions or may be an Application Specific Integrated Circuit. Computer instructions may be stored on storage medium (such as the system memory 740). The computer instructions may be embodied as computer software that contains computer readable code which, when loaded onto the system processing unit 740, configures the system processing unit 740 to perform a method according to embodiments of the disclosure.

[0089] The edge system 700 comprises an output unit 760 configured to output an indication of the recognised objects from the one or more images captured by the first SOI image capture unit 720 and / or the second SOI image capture unit 710. The indication of the recognised objects may be provided to the output unit 760 by the system processing unit 740.

[0090] The output unit 760 may comprise a display unit (not shown) configured to display the indication of the recognised objects from the one or more images captured by the first SOI image capture unit 720 and / or the second SOI image capture unit 710. The output unit 760 may comprise a speaker unit (not shown) configured to speak the indication of the recognised objects from the one or more images captured by the first SOI image capture unit 720 and / or the second SOI image capture unit 710. The output unit 760 may comprise wireless transmission unit configured to transmit the indication of the recognised objects from the one or more images captured by the first SOI image capture unit 720 and / or the second SOI image capture unit 710 via a wireless communications network. If the edge system is worn by a rescue worker (in a similar way to a body cam), the rescue worker may use the indication of the recognised objects from the one or more images captured by the first SOI image capture unit 720 and / or the second SOI to identify and / or locate injured persons, or understand the extent of an emergency quickly. For example, the indication may comprise the metadata previously explained.

[0091] In some embodiments, the edge system 700 comprises a system memory 730 configured to store the images captured by the first SOI image capture unit 720 and the images captured by the second SOI image capture unit 710. In some embodiments, the system memory 730 stores the first and second Al object recognition models. The system processing unit 740 may be configured to control the system memory 730 to delete images after they have been processed for object recognition.

[0092] Figure 6 illustrates an alternative configuration of an edge system 800 in accordance with example embodiments. The edge system 800 may be wearable. Alternatively, the edge system 800 may be comprised in or mounted on a vehicle.

[0093] The edge system 800 comprises a plurality of site-of-interest (SOI) cameras (comprising a first SOI camera 820 and a second SOI camera 810).

[0094] The first SOI camera 820 comprises the first SOI image capture unit 720 described with reference to Figure 5. The first SOI camera 820 also comprises a first SOI camera memory 822 and a first SOI camera processing unit 824. The first SOI camera processing unit 824 is configured to apply a first Al object recognition model to the one or more images captured by the first SOI image capture unit 821 to recognise one or more objects in the SOI. The first SOI camera processing unit 824 may be configured to retrieve the images captured by the first SOI image capture unit 720 from the first SOI image capture unit 720, or from the first SOI camera memory 822.

[0095] The first SOI camera processing unit 824 may be a microprocessor carrying out computer instructions or may be an Application Specific Integrated Circuit. Computer instructions may be stored on storage medium (such as the first SOI camera memory 822). The computer instructions may be embodied as computer software that contains computer readable code which, when loaded onto the first SOI camera processing unit 824, configures the first SOI camera processing unit 824 to perform a method according to embodiments of the disclosure.

[0096] In some embodiments, the first SOI camera memory 822 is configured to store the images captured by the first SOI image capture unit 720. In some embodiments, the first SOI camera memory 822 also stores the first Al object recognition model.

[0097] The second SOI camera 810 comprises the second SOI camera processing unit 710 described with reference to Figure 5. The second SOI camera 810 also comprises a second SOI camera memory 812 and a second SOI camera processing unit 814.

[0098] The second SOI camera processing unit 814 is configured to apply a second Al object recognition model to the one or more images captured by the second SOI image capture unit 710 to recognise one or more objects in the SOI. The first and second Al object recognitions model are adapted for recognising the objects in different types of SOI.

[0099] The second SOI camera processing unit 814 may be configured to retrieve the images captured by the second SOI image capture unit 710 from the second SOI image capture unit 710, or from the second SOI camera memory 812.

[0100] The second SOI camera processing unit 814 may be a microprocessor carrying out computer instructions or may be an Application Specific Integrated Circuit. Computer instructions may be stored on storage medium (such as the second SOI camera memory 812). The computer instructions may be embodied as computer software that contains computer readable code which, when loaded onto the second SOI camera processing unit 814, configures the second SOI camera processing unit 814 to perform a method according to embodiments of the disclosure.

[0101] In some embodiments, the second SOI image capture unit 710 comprises the second SOI camera memory 812. The second SOI camera memory 812 is configured to store the images captured by the second image capture unit 710. In some embodiments, the second SOI camera memory 812 stores the second Al object recognition model.

[0102] The edge system 800 comprises an output unit 760 configured to output an indication of the recognised objects from the one or more images captured by the first SOI image capture unit 720 and / or the second SOI image capture unit 710. The output unit 760 has already been described with reference to Figure 5 and that description will not be repeated here for brevity.

[0103] In some embodiments, the first SOI camera processing unit 824 may provide an indication of the recognised objects from the one or more images captured by the first SOI image capture unit 720 to the output unit 760 and the second SOI camera processing unit 814 may provide an indication of the recognised objects from the one or more images captured by the second SOI image capture unit 710 to the output unit 760.

[0104] Therefore, whereas in Figure 5, the edge system 700 comprises a system memory 730 and system processing unit 740 for the first SOI image capture unit 720 and the second SOI camera 710, the edge system 800 in Figure 6 comprises a dedicated first SOI camera memory 822 and dedicated first SOI camera processing unit 824 in the first SOI camera 820 and a dedicated second SOI camera memory 812 and dedicated second SOI camera processing unit 814 in the second SOI camera 810.

[0105] Each of the memories 812, 822, 730 and may be referred to as “memory”.

[0106] Each of the processing units 740, 814, 824 com may be referred to as” processing circuitry”.

[0107] Figure 7 illustrates a method of operating an edge system for a vehicle in accordance with example embodiments. The method starts in step S1.

[0108] In step S2, the method comprises capturing, by a drive path image capture unit comprised in the edge system, one or more images of a drive path of the vehicle.

[0109] In step S3, the method comprises capturing, by a site-of-interest (SOI) image capture unit comprised in the edge system, one or more images of an SOI monitored by the vehicle.

[0110] In step S4, the method comprises applying, by processing circuitry comprised in the edge system, a first Al object recognition model to the one or more images captured by the drive path camera to recognise one or more objects in the drive path of the vehicle. The first Al recognition model is adapted for recognising the objects in the drive path of the vehicle.

[0111] In step S5, the method comprises applying, by the processing circuitry comprised in the edge system, a second Al object recognition model to the one or more images captured by the SOI camera to recognise one or more objects in the SOI. The second Al object recognition model is adapted for recognising the objects in the SOI.

[0112] In step S6, the method comprises providing, by the processing circuitry comprised in the edge system, an indication of the recognised objects in the drive path of the vehicle to a drive control unit for the vehicle.

[0113] In step S7, the method comprises providing, by the processing circuitry comprised in the edge system, an indication of the recognised objects in the SOI to a wireless transmission unit comprised in the edge system.

[0114] In step S8, the method comprises transmitting, by the wireless transmission unit, the indication of the recognised objects in the SOI via a wireless communications network.

[0115] In step S9, the method ends.

[0116] Figure 8 illustrates a method of operating an edge system in accordance with example embodiments. The method starts in step S10.

[0117] In step S11 , the method comprises capturing, by each of a plurality of site-of-interest (SOI) image capture units comprised in the edge system, one or more images of an SOI.

[0118] In step S12, the method comprises applying, by processing circuitry comprised in the edge system, a first Al object recognition model to the one or more images captured by a first of the plurality of SOI image capture units to recognise one or more objects in the SOI. In step S13, the method comprises applying, by the processing circuitry, a second Al object recognition model to the one or more images captured by a second of the plurality of SOI image capture units to recognise one or more objects in the SOI. The first and second Al object recognition models are adapted for recognising the objects in different types of SOI.

[0119] In step S14, the method comprises outputting, by an output unit of the edge system, an indication of the recognised objects from the one or more images captured by the first SOI image capture unit and / or the second SOI image capture unit.

[0120] In step S16, the method ends.

[0121] Embodiments of the present disclosure are not particularly limited to the number and arrangement of method steps which have been illustrated with reference to the specific example of Figures 7 and 8 of the present disclosure. Any number of additional steps may be provided. Alternatively, or in addition, at least one or more of the steps of the method of Figure 7 or Figure 8 may be performed in sequence and / or in parallel with the other steps of the method. Hence, the present disclosure is not particularly limited in this regard.

[0122] The following numbered paragraphs provide further example aspects and features of the present technique:

[0123] Paragraph 1. An edge system for a vehicle, the edge system comprising a drive path image capture unit configured to capture one or more images of a drive path of the vehicle, a site-of-interest (SOI) camera configured to capture one or more images of an SOI monitored by the vehicle; processing circuitry configured to apply a first artificial intelligence (Al) object recognition model to the one or more images captured by the drive path image capture unit to recognise one or more objects in the drive path of the vehicle, the first Al recognition model being adapted for recognising the objects in the drive path of the vehicle, apply a second Al object recognition model to the one or more images captured by the SOI image capture unit to recognise one or more objects in the SOI, the second Al object recognition model being adapted for recognising the objects in the SOI; provide an indication of the recognised objects in the drive path of the vehicle to a drive control unit for the vehicle; provide an indication of the recognised objects in the SOI to a wireless transmission unit comprised in the edge system, the wireless transmission unit being configured to transmit the indication of the recognised objects in the SOI via a wireless communications network.

[0124] Paragraph 2. An edge system according to paragraph 1 , wherein edge system comprises memory configured to store the images captured by the drive path image capture unit and the SOI image capture unit, wherein the processing circuitry is configured to transmit an instruction to the memory to delete the images captured by the drive path image capture unit after the first Al object recognition model has been applied to the images captured by the drive path image capture unit, and delete the images captured by the SOI image capture unit after the second Al object recognition model has been applied to the images captured by the SOI image capture unit.

[0125] Paragraph 3. An edge system according to paragraph 1 or paragraph 2, wherein the indication of the recognised objects in the SOI is metadata of the recognised objects in the SOI. Paragraph 4. An edge system according to paragraph 3, wherein the metadata comprises one or more of a number of recognised objects in the SOI, a number of persons in the SOI, a number of vehicles in the SOI, an indication of a state of the persons or the vehicles in the SOI.

[0126] Paragraph 5. An edge system according to any of paragraphs 1 to 4, wherein the edge system comprises a spotlight projection unit configured to project one or more spotlights onto one or more of the objects recognised in the SOI.

[0127] Paragraph 6. An edge system according to paragraph 5, wherein the one or more spotlights are coloured LED spotlights.

[0128] Paragraph 7. An edge system according to paragraph 5, wherein the one or more spotlights are laser spot lights.

[0129] Paragraph 8. An edge system according to paragraph 7, wherein laser spot lights are infrared laser spot lights.

[0130] Paragraph 9. An edge system according to any of paragraphs 1 to 8, wherein the vehicle is an unmanned aerial vehicle (UAV).

[0131] Paragraph 10. An edge system according to any of paragraphs 1 to 8, wherein the vehicle is a terrestrial vehicle.

[0132] Paragraph 11. An edge system according to any of paragraphs 1 to 10, wherein the edge system comprises one or more additional SOI image capture units each configured to capture one or more images of the SOI, wherein the processing circuitry is configured to apply one or more additional Al object recognition models to the images captured by the one or more additional SOI image capture units respectively, the second Al object recognition model and the one or more additional Al object recognition models forming a plurality of Al object recognition models each adapted for recognising objects in different types of SOI.

[0133] Paragraph 12. A vehicle comprising an edge system according to any of paragraphs 1 to 11. Paragraph 13. An edge system, the edge system comprising a plurality of site-of-interest (SOI) cameras each configured to capture one or more images of an SOI, processing circuitry configured to apply a first artificial intelligence (Al) object recognition model to the one or more images captured by a first of the plurality of SOI image capture units to recognise one or more objects in the SOI, apply a second Al object recognition model to the one or more images captured by a second of the plurality of SOI image capture units to recognise one or more objects in the SOI, the first and second Al object recognitions model being adapted for recognising the objects in different types of SOI; and an output unit configured to output an indication of the recognised objects from the one or more images captured by the first SOI image capture unit and / or the second SOI image capture unit.

[0134] Paragraph 14. An edge system according to paragraph 13, wherein the edge system is wearable by a person.

[0135] Paragraph 15. An edge system according to paragraph 13 or paragraph 14, wherein the output unit comprises a display unit configured to display the indication of the recognised objects from the one or more images captured by the first SOI image capture unit and / or the second SOI image capture unit

[0136] Paragraph 16. An edge system according to any of paragraphs 13 to 15, wherein the output unit comprises a speaker unit configured to speak the indication of the recognised objects from the one or more images captured by the first SOI image capture unit and / or the second SOI image capture unit.

[0137] Paragraph 17. An edge system according to any of paragraphs 13 to 16, wherein the output unit comprises a wireless transmission unit configured to transmit the indication of the recognised objects from the one or more images captured by the first SOI image capture unit and / or the second SOI image capture unit via a wireless communications network.

[0138] Paragraph 18. A method of operating an edge system for a vehicle, the method comprising capturing, by a drive path image capture unit comprised in the edge system, one or more images of a drive path of the vehicle, capturing, by a site-of-interest (SOI) image capture units comprised in the edge system, one or more images of an SOI monitored by the vehicle; applying, by processing circuitry comprised in the edge system, a first artificial intelligence (Al) object recognition model to the one or more images captured by the drive path image capture unit to recognise one or more objects in the drive path of the vehicle, the first Al recognition model being adapted for recognising the objects in the drive path of the vehicle, applying, by the processing circuitry comprised in the edge system, a second Al object recognition model to the one or more images captured by the SOI image capture unit to recognise one or more objects in the SOI, the second Al object recognition model being adapted for recognising the objects in the SOI; providing, by the processing circuitry comprised in the edge system, an indication of the recognised objects in the drive path of the vehicle to a drive control unit for the vehicle; providing, by the processing circuitry comprised in the edge system, an indication of the recognised objects in the SOI to a wireless transmission unit comprised in the edge system, transmitting, by the wireless transmission unit, the indication of the recognised objects in the SOI via a wireless communications network.

[0139] Paragraph 19. A method of operating an edge system, the method comprising capturing, by each of a plurality of site-of-interest (SOI) image capture units comprised in the edge system, one or more images of an SOI, applying, by processing circuitry comprised in the edge system, a first Al object recognition model to the one or more images captured by a first of the plurality of SOI image capture units to recognise one or more objects in the SOI, applying, by the processing circuitry, a second Al object recognition model to the one or more images captured by a second of the plurality of SOI image capture units to recognise one or more objects in the SOI, the first and second Al object recognitions model being adapted for recognising the objects in different types of SOI; outputting, by an output unit of the edge system, an indication of the recognised objects from the one or more images captured by the first SOI image capture unit and / or the second SOI image capture unit.

[0140] Paragraph 20. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to perform the method of any of paragraphs 18 to 19.

[0141] Paragraph 21. A non-transitory computer-readable storage medium storing a computer program comprising instructions according to paragraph 20.

[0142] It will be appreciated that the above description for clarity has described embodiments with reference to different functional units, circuitry and / or processors. However, it will be apparent that any suitable distribution of functionality between different functional units, circuitry and / or processors may be used without detracting from the embodiments. Described embodiments may be implemented in any suitable form including hardware, software, firmware or any combination of these. Described embodiments may optionally be implemented at least partly as computer software running on one or more data processors and / or digital signal processors. The elements and components of any embodiment may be physically, functionally and logically implemented in any suitable way. Indeed, the functionality may be implemented in a single unit, in a plurality of units or as part of other functional units. As such, the disclosed embodiments may be implemented in a single unit or may be physically and functionally distributed between different units, circuitry and / or processors. Although the present disclosure has been described in connection with some embodiments, it is not intended to be limited to the specific form set forth herein. Additionally, although a feature may appear to be described in connection with particular embodiments, one skilled in the art would recognise that various features of the described embodiments may be combined in any manner suitable to implement the technique.

Claims

CLAIMS1 . An edge system for a vehicle, the edge system comprising a drive path image capture unit configured to capture one or more images of a drive path of the vehicle, a site-of-interest (SOI) camera configured to capture one or more images of an SOI monitored by the vehicle; processing circuitry configured to apply a first artificial intelligence (Al) object recognition model to the one or more images captured by the drive path image capture unit to recognise one or more objects in the drive path of the vehicle, the first Al recognition model being adapted for recognising the objects in the drive path of the vehicle, apply a second Al object recognition model to the one or more images captured by the SOI image capture unit to recognise one or more objects in the SOI, the second Al object recognition model being adapted for recognising the objects in the SOI; provide an indication of the recognised objects in the drive path of the vehicle to a drive control unit for the vehicle; provide an indication of the recognised objects in the SOI to a wireless transmission unit comprised in the edge system, the wireless transmission unit being configured to transmit the indication of the recognised objects in the SOI via a wireless communications network.

2. An edge system according to claim 1 , wherein edge system comprises memory configured to store the images captured by the drive path image capture unit and the SOI image capture unit, wherein the processing circuitry is configured to transmit an instruction to the memory to delete the images captured by the drive path image capture unit after the first Al object recognition model has been applied to the images captured by the drive path image capture unit, and delete the images captured by the SOI image capture unit after the second Al object recognition model has been applied to the images captured by the SOI image capture unit.

3. An edge system according to claim 1 , wherein the indication of the recognised objects in the SOI is metadata of the recognised objects in the SOI.

4. An edge system according to claim 3, wherein the metadata comprises one or more of a number of recognised objects in the SOI, a number of persons in the SOI, a number of vehicles in the SOI, an indication of a state of the persons or the vehicles in the SOI.

5. An edge system according to claim 1 , wherein the edge system comprisesa spotlight projection unit configured to project one or more spotlights onto one or more of the objects recognised in the SOI.

6. An edge system according to claim 5, wherein the one or more spotlights are coloured LED spotlights.

7. An edge system according to claim 5, wherein the one or more spotlights are laser spot lights.

8. An edge system according to claim 7, wherein laser spot lights are infrared laser spot lights.

9. An edge system according to claim 1 , wherein the vehicle is an unmanned aerial vehicle (UAV).

10. An edge system according to claim 1 , wherein the vehicle is a terrestrial vehicle.

11. An edge system according to claim 1 , wherein the edge system comprises one or more additional SOI image capture units each configured to capture one or more images of the SOI, wherein the processing circuitry is configured to apply one or more additional Al object recognition models to the images captured by the one or more additional SOI image capture units respectively, the second Al object recognition model and the one or more additional Al object recognition models forming a plurality of Al object recognition models each adapted for recognising objects in different types of SOI.

12. A vehicle comprising an edge system according to claim 1 .

13. An edge system, the edge system comprising a plurality of site-of-interest (SOI) cameras each configured to capture one or more images of an SOI, processing circuitry configured to apply a first artificial intelligence (Al) object recognition model to the one or more images captured by a first of the plurality of SOI image capture units to recognise one or more objects in the SOI, apply a second Al object recognition model to the one or more images captured by a second of the plurality of SOI image capture units to recognise one or more objects in the SOI,the first and second Al object recognitions model being adapted for recognising the objects in different types of SOI; and an output unit configured to output an indication of the recognised objects from the one or more images captured by the first SOI image capture unit and / or the second SOI image capture unit.

14. An edge system according to claim 13, wherein the edge system is wearable by a person.

15. An edge system according to claim 13, wherein the output unit comprises a display unit configured to display the indication of the recognised objects from the one or more images captured by the first SOI image capture unit and / or the second SOI image capture unit16. An edge system according to claim 13, wherein the output unit comprises a speaker unit configured to speak the indication of the recognised objects from the one or more images captured by the first SOI image capture unit and / or the second SOI image capture unit.

17. An edge system according to claim 13, wherein the output unit comprises a wireless transmission unit configured to transmit the indication of the recognised objects from the one or more images captured by the first SOI image capture unit and / or the second SOI image capture unit via a wireless communications network.

18. A method of operating an edge system for a vehicle, the method comprising capturing, by a drive path image capture unit comprised in the edge system, one or more images of a drive path of the vehicle, capturing, by a site-of-interest (SOI) image capture units comprised in the edge system, one or more images of an SOI monitored by the vehicle; applying, by processing circuitry comprised in the edge system, a first artificial intelligence (Al) object recognition model to the one or more images captured by the drive path image capture unit to recognise one or more objects in the drive path of the vehicle, the first Al recognition model being adapted for recognising the objects in the drive path of the vehicle, applying, by the processing circuitry comprised in the edge system, a second Al object recognition model to the one or more images captured by the SOI image capture unit to recognise one or more objects in the SOI, the second Al object recognition model being adapted for recognising the objects in the SOI; providing, by the processing circuitry comprised in the edge system, an indication of the recognised objects in the drive path of the vehicle to a drive control unit for the vehicle; providing, by the processing circuitry comprised in the edge system, an indication of the recognised objects in the SOI to a wireless transmission unit comprised in the edge system,transmitting, by the wireless transmission unit, the indication of the recognised objects in the SOI via a wireless communications network.

19. A method of operating an edge system, the method comprising capturing, by each of a plurality of site-of-interest (SOI) image capture units comprised in the edge system, one or more images of an SOI, applying, by processing circuitry comprised in the edge system, a first Al object recognition model to the one or more images captured by a first of the plurality of SOI image capture units to recognise one or more objects in the SOI, applying, by the processing circuitry, a second Al object recognition model to the one or more images captured by a second of the plurality of SOI image capture units to recognise one or more objects in the SOI, the first and second Al object recognitions model being adapted for recognising the objects in different types of SOI; outputting, by an output unit of the edge system, an indication of the recognised objects from the one or more images captured by the first SOI image capture unit and / or the second SOI image capture unit.

20. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to perform the method of claim 18 or claim 19.

21. A non-transitory computer-readable storage medium storing a computer program comprising instructions according to claim 20.

Images