Dynamic slot management method for cluster unmanned aerial vehicle multi-source heterogeneous data flow

Abstract

The invention discloses a dynamic slot management method and device for cluster unmanned aerial vehicle multi-source heterogeneous data streams, and belongs to the technical field of unmanned aerial vehicle data analysis. The method comprises the following steps: receiving multi-source heterogeneous data streams of cluster unmanned aerial vehicles, and classifying and distributing unique stream identifiers according to video data, SAR radar data and target data; searching the existing data processing slots bound with the same identifier, directly using the existing data processing slots if the existing data processing slots exist, allocating idle slots if the existing data processing slots do not exist, and executing a timestamp-based LRU elimination strategy or dynamically adjusting slot allocation according to data stream characteristics when the slots are fully occupied; and after the data frames are distributed to the corresponding slot positions, updating the recent access timestamps of the slot positions as the current system time. According to the method, intelligent resource allocation and recovery of multi-source heterogeneous data streams are realized, the utilization efficiency of system resources is improved, active data streams are guaranteed to be processed preferentially, and the real-time performance and stability of an unmanned aerial vehicle cluster control system are enhanced.

Description

Technical Field

[0001] This application relates to the field of UAV data analysis technology, and in particular to a dynamic slot management method and apparatus for multi-source heterogeneous data streams of swarm UAVs. Background Technology

[0002] In UAV swarm control systems, concurrent processing of multiple heterogeneous data streams often presents the technical problem of low resource management efficiency. Existing technologies require UAV systems to simultaneously process various heterogeneous data streams, such as visible light video streams, infrared video streams, and radar data. Traditional methods typically allocate fixed resources to each data stream or employ a simple first-in-first-out (FIFO) elimination strategy, resulting in low system resource utilization, untimely processing of active data streams, and impacting the real-time situational awareness and decision-making capabilities of the UAV swarm. Summary of the Invention

[0003] The main objective of this application is to provide a dynamic slot management method and apparatus for multi-source heterogeneous data streams of swarmed UAVs. The aim is to achieve intelligent resource allocation and recycling of multi-source heterogeneous data streams through a unified slot management architecture and a timestamp-based LRU eviction mechanism, thereby improving system resource utilization efficiency, ensuring priority processing of active data streams, and thus enhancing the real-time performance and stability of the UAV swarm control system.

[0004] To achieve the above objectives, this application provides a dynamic slot management method for multi-source heterogeneous data streams from swarmed UAVs, comprising: The system receives multi-source heterogeneous data streams from a cluster of UAVs and assigns a unique stream identifier to each data stream. The data streams are classified according to their data types, which include video data, SAR radar data, and target data. The system checks if any existing data processing slots exist and are bound to the same data stream identifier. If they do, the data processing slot is used directly. If not, and there are available data processing slots, an available data processing slot is allocated to the data stream. If all data processing slots are occupied, a timestamp-based LRU eviction policy is executed, and the data frame is finally distributed to the corresponding data processing slot. Here, each data processing slot is bound to a data stream identifier, the most recently accessed timestamp, and the corresponding processing worker thread in a one-to-one correspondence. Alternatively, the data processing slot allocation strategy can be dynamically adjusted based on the characteristics of the data stream. Update the most recently accessed timestamp of this data processing slot to the current system time.

[0005] Optionally, the execution of the timestamp-based LRU eviction policy includes: Initialize the LRU index and minimum timestamp; Iterate through all data processing slots and compare the most recent access timestamps of each data processing slot. Determine the data processing slot with the smallest timestamp as the data processing slot to be eliminated. Rebind the data processing slot to be phased out to a new stream identifier and update its most recently accessed timestamp to the current timestamp.

[0006] Optionally, before receiving the multi-source heterogeneous data stream from the cluster of UAVs, the method further includes: Initialize the corresponding slot pool for each data stream distributor to obtain a slot pool where all data processing slots are idle; The number of data processing slots in the slot pool is fixed and determined based on the data processing capacity. The type of the slot pool corresponds one-to-one with the data type.

[0007] Optionally, the step of checking whether there is a data processing slot with the same data flow identifier among the existing data processing slots includes: The new data stream is assigned to the corresponding slot pool based on its unique identifier. Then, search the slot pool for a data processing slot that is bound to the same flow identifier.

[0008] Optionally, the step of traversing all data processing slots and comparing the most recent access timestamps of each data processing slot to determine the data processing slot with the smallest timestamp as the data processing slot to be phased out includes: Compare the timestamp of the current data processing slot with the smallest timestamp that has been recorded; If the timestamp of the current data processing slot is smaller, update the minimum timestamp to the timestamp of the current data processing slot and record the index of that slot; After the traversal is complete, the slot corresponding to the recorded slot index is used as the elimination target.

[0009] Optionally, after updating the most recently accessed timestamp of the data processing slot to the current system time, the method further includes: Data frames from multi-source heterogeneous data streams are processed, including video decoding, image decoding, and target parameter value parsing. Update the most recently accessed timestamp to the current system timestamp.

[0010] Optionally, after allocating an idle data processing slot to the data stream, the method further includes: Synchronize and update the timestamps of the data processing slots.

[0011] Optionally, the method further includes: Regularly check the usage status of data processing slots, and automatically unbind and mark data processing slots that have not received data for longer than the preset timeout threshold as idle.

[0012] Optionally, a data processing slot allocation strategy can be dynamically adjusted based on data flow characteristics, including: Monitor the processing queue length and processing latency of each data processing slot; When data processing congestion is detected, the queue size is dynamically adjusted according to the data stream characteristics, and the allocation of data processing slots in the data stream is dynamically adjusted according to the queue size.

[0013] To achieve the above objectives, this application also provides a dynamic slot management device for multi-source heterogeneous data streams from swarmed UAVs, comprising: The identifier allocation module is used to receive multi-source heterogeneous data streams from a cluster of UAVs and assign a unique stream identifier to each data stream. The data streams are classified according to their data types, which include video data, SAR radar data, and target data. The data distribution module checks if any existing data processing slots exist and are bound to the same data stream identifier. If they do, the data processing slot is used directly. If not, and there are available data processing slots, an available data processing slot is allocated to the data stream. If all data processing slots are occupied, a timestamp-based LRU eviction policy is executed to distribute the data frame to the corresponding data processing slot. Each data processing slot is bound to a data stream identifier, a recent access timestamp, and a corresponding processing worker thread. Alternatively, the data processing slot allocation strategy can be dynamically adjusted based on the data stream characteristics. The update module is used to update the most recently accessed timestamp of the data processing slot to the current system time.

[0014] This application proposes a dynamic slot management method and apparatus for multi-source heterogeneous data streams from swarmed UAVs. The method includes receiving multi-source heterogeneous data streams from swarmed UAVs and assigning a unique stream identifier to each data stream. The data streams are categorized according to data type, including video data, SAR radar data, and target data. The method searches existing data processing slots to see if a data processing slot with the same data stream identifier exists. If it does, the slot is used directly. If it does not exist and there are available data processing slots, an available slot is assigned to the data stream. If all data processing slots are occupied, a timestamp-based LRU eviction policy is executed, ultimately distributing data frames to the corresponding data processing slots. Each data processing slot is bound to a data stream identifier, a recent access timestamp, and a corresponding processing worker thread. Alternatively, the data processing slot allocation strategy can be dynamically adjusted based on data stream characteristics. The recent access timestamp of the data processing slot is updated to the current system time. This invention constructs a unified slot management framework, achieving effective management of heterogeneous data streams such as video streams, infrared data, radar images, and target information through classification identification and differentiated processing strategies. Based on the LRU algorithm and timestamp mechanism, this invention prioritizes the retention and processing of active data streams, reducing the risk of critical data loss, improving system reliability, and solving the problem of dynamic slot management for multi-source heterogeneous data streams in swarm UAVs. It achieves reduced processing latency and improved system response speed through intelligent resource allocation and recycling mechanisms, ensuring the stable operation of the UAV swarm control system. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating an embodiment of the dynamic slot management method for multi-source heterogeneous data streams of clustered UAVs provided in this application.

[0016] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0018] This invention proposes a dynamic slot management method for multi-source heterogeneous data streams of swarm UAVs. Through a unified slot management architecture and a timestamp-based LRU eviction mechanism, it realizes intelligent resource allocation and recycling of multi-source heterogeneous data streams, improves system resource utilization efficiency, ensures priority processing of active data streams, and thus enhances the real-time performance and stability of the UAV swarm control system.

[0019] Specifically, the present invention aims to solve the following technical problems: 1. Inefficient resource allocation: Traditional fixed slot allocation methods cannot dynamically adjust resources according to the actual needs of the data flow, leading to resource waste or shortage. This invention improves resource utilization efficiency by using a dynamic slot allocation mechanism to intelligently allocate processing resources based on the activity level and priority of the data flow.

[0020] 2. Challenges in Unified Management of Heterogeneous Data Streams: The multi-source heterogeneous data streams generated by UAVs possess different data characteristics, processing requirements, and timeliness needs, making unified and efficient management difficult using traditional methods. This invention constructs a unified slot management framework, achieving effective management of heterogeneous data streams such as video streams, infrared data, radar images, and target information through classification, labeling, and differentiated processing strategies.

[0021] 3. Risk of Critical Data Loss: In resource-constrained environments, traditional methods often cannot guarantee the timely processing of critical data streams, potentially leading to the loss of important information. This invention, based on the LRU algorithm and timestamp mechanism, prioritizes the retention and processing of active data streams, reducing the risk of critical data loss and improving system reliability.

[0022] 4. Insufficient system real-time performance and stability: UAV control systems have extremely high requirements for the real-time performance and stability of data processing, which traditional methods struggle to meet. This invention reduces processing latency and improves system response speed through intelligent resource allocation and recycling mechanisms, ensuring the stable operation of the UAV swarm control system.

[0023] Reference Figure 1 The dynamic slot management method for multi-source heterogeneous data streams of swarmed UAVs provided in the first embodiment of this application may include: S100: Initialize the corresponding slot pool for each data stream distributor to obtain a slot pool where all data processing slots are idle. The number of data processing slots in the slot pool is fixed and determined based on the data processing capacity. The type of the slot pool corresponds one-to-one with the data type.

[0024] Specifically, during system initialization, slot pools are created for different types of data streams, corresponding to video data, SAR radar data, and target data, respectively: a video data slot pool, a SAR radar data slot pool, and a target data slot pool. For example, each slot pool initializes four data processing slots based on the system's processing capacity. Each data processing slot contains a data stream identifier, a recent access timestamp, and a corresponding processing worker thread. Initially, all data processing slots are idle, waiting to be bound to a data stream.

[0025] Clearly, by dividing the slot pool according to data type and initializing a fixed number of slots, it is possible to classify and manage different types of data streams, avoid mutual interference between different types of data streams during resource allocation, lay the foundation for efficient slot allocation and data processing in the future, and improve the targeting of system resource allocation.

[0026] S101. Receive multi-source heterogeneous data streams from a cluster of UAVs and assign a unique stream identifier to each data stream. The data streams are classified according to their data types, which include: video data, SAR radar data, and target data. The multi-source heterogeneous data streams are transmitted to the system via UDP multicast, originating from multiple drones in the cluster. Specifically, a unique stream identifier, the drone ID, is assigned to each data stream. This identifier is used for subsequent data processing slot allocation and management, ensuring that each data stream can be accurately identified and tracked.

[0027] Understandably, receiving data streams via UDP multicast can meet the needs of concurrent transmission of multi-source data from clustered drones. Using the drone number as a unique stream identifier, combined with data type classification, can quickly distinguish data streams from different sources and of different types, solving the problem of difficult unified identification and management of heterogeneous data streams, and providing an accurate identification basis for subsequent dynamic slot allocation.

[0028] S102. Check if there is a data processing slot with the same data stream identifier among the existing data processing slots. If it exists, use the data processing slot directly. If it does not exist and there is an empty data processing slot, allocate an empty data processing slot for the data stream. If all data processing slots are occupied, execute the timestamp-based LRU eviction policy and finally distribute the data frame to the corresponding data processing slot. The data processing slot is bound one-to-one with the data stream identifier, the most recent access timestamp, and the corresponding processing worker thread. Alternatively, the data processing slot allocation strategy can be dynamically adjusted based on the data stream characteristics. In one embodiment of this application, the step of checking whether there is a data processing slot bound to the same data flow identifier among existing data processing slots includes: The new data stream is assigned to the corresponding slot pool based on its unique identifier. Then, search the slot pool for a data processing slot that is bound to the same flow identifier.

[0029] Specifically, when a new data stream arrives, the corresponding slot pool is first determined according to its data type (e.g., video data corresponds to the video data slot pool), and then the data processing slot that has been bound to the unique identifier (drone number) of the data stream is searched in the slot pool.

[0030] Clearly, this method of first dividing the data into pools and then searching can narrow the search range, reduce search time, improve the efficiency of slot search, ensure that the data stream can quickly match the corresponding processing slot, and improve the system's response speed.

[0031] In one embodiment of this application, executing a timestamp-based LRU eviction policy includes: Initialize the LRU index and minimum timestamp; Iterate through all data processing slots and compare the most recent access timestamps of each data processing slot. Determine the data processing slot with the smallest timestamp as the data processing slot to be eliminated. Rebind the data processing slot to be phased out to a new stream identifier and update its most recently accessed timestamp to the current timestamp.

[0032] The core of the LRU eviction policy is to prioritize the eviction of the least recently used data processing slots, ensuring that active data streams receive sufficient processing resources. Specifically, the LRU index and minimum timestamp are set during initialization to prepare for subsequent timestamp comparisons.

[0033] Specifically, by rebinding the slots to be phased out with new stream identifiers and updating the timestamps, the recycling of data processing slots is achieved, avoiding resource waste, while ensuring that new data streams can obtain processing resources in a timely manner, thus solving the problem of insufficient resources under the traditional fixed resource allocation method.

[0034] In one embodiment of this application, the step of traversing all data processing slots and comparing the most recent access timestamps of each data processing slot to determine the data processing slot with the smallest timestamp as the data processing slot to be phased out includes: Compare the timestamp of the current data processing slot with the smallest timestamp that has been recorded; If the timestamp of the current data processing slot is smaller, update the minimum timestamp to the timestamp of the current data processing slot and record the index of that slot; After the traversal is complete, the slot corresponding to the recorded slot index is used as the elimination target.

[0035] Specifically, during the traversal, the timestamp of the first data processing slot is first used as the initial minimum timestamp. Then, the timestamp of each subsequent slot is compared with the current minimum timestamp. If the timestamp of a subsequent slot is smaller, the minimum timestamp is updated and the index information of that slot is recorded. For example, taking a slot pool with 4 slots as an example, after traversing the 4 slots, the slot corresponding to the finally recorded index is the least recently used slot.

[0036] In fact, this traversal comparison method can accurately identify the least used slots, ensuring the precise execution of the LRU eviction policy, making resource recycling and reallocation more targeted, and further improving resource utilization efficiency.

[0037] In one embodiment of this application, after allocating an idle data processing slot to the data stream, the method further includes: Synchronize and update the timestamps of the data processing slots.

[0038] Specifically, once an empty slot is allocated to a new data stream, the most recent access timestamp of that slot is immediately updated to the current system timestamp to ensure that the data stream corresponding to that slot is marked as active.

[0039] It can be said without a doubt that updating the timestamp synchronously after allocating slots can prevent the active data stream from being mistakenly eliminated due to the timestamp not being updated in time, ensuring that the data stream can continuously obtain processing resources and guaranteeing the continuity and stability of data processing.

[0040] S103. Update the most recently accessed timestamp of this data processing slot to the current system time.

[0041] In one embodiment of this application, after updating the most recently accessed timestamp of the data processing slot to the current system time, the method further includes: Data frames from multi-source heterogeneous data streams are processed, including video decoding, image decoding, and target parameter value parsing. Update the most recently accessed timestamp to the current system timestamp.

[0042] Once a data frame is distributed to its corresponding processing slot, the processing worker thread performs corresponding processing operations on the data frame. For example, it performs video decoding on video data frames, image decoding on SAR radar data frames, and target parameter value parsing on target data frames. Specifically, the system employs a feedback mechanism: whenever a slot successfully processes a data frame, it immediately updates the slot's most recent access time to the current system timestamp, forming a dual update mechanism.

[0043] Clearly, by processing data frames in a targeted manner, the processing needs of different types of data streams can be met, while the dual update mechanism ensures that the timestamp of active data streams always reflects their latest activity status, further reducing the risk of active data streams being mistakenly eliminated and improving the reliability of data processing.

[0044] In one embodiment of this application, the method further includes: Regularly check the usage status of data processing slots, and automatically unbind and mark data processing slots that have not received data for longer than the preset timeout threshold as idle.

[0045] Specifically, the preset timeout threshold is 5 minutes. The system will periodically check the last access timestamp of each data processing slot. If the last access timestamp of a slot is more than 5 minutes away from the current time and no data from the corresponding data stream is received during the period, the slot will be automatically unbound from the original data stream identifier and the slot will be marked as idle.

[0046] For example, this step can promptly release long-unused resources, reserve processing capacity for new data streams, avoid resources being idle for a long time, further optimize the utilization rate of system resources, and ensure that the system always maintains an efficient resource allocation state during long-term operation.

[0047] Furthermore, in another embodiment of this application, the strategy of dynamically adjusting the data processing slot allocation based on the data flow characteristics specifically includes: monitoring the processing queue length and processing latency of each data processing slot, and dynamically adjusting the queue size according to the data flow characteristics when congestion is detected, thereby realizing intelligent scheduling of the data flow.

[0048] During data processing, the system monitors the processing queue length (i.e., the number of data frames waiting to be processed in that slot) and processing latency (i.e., the time from when a data frame enters the slot to when processing is completed) of each data processing slot in real time. Specifically, when the processing queue length of a slot exceeds a preset threshold, or the processing latency exceeds a preset latency threshold, the slot is determined to be congested. For example, for video data streams, which have a large data volume and long processing time, when the corresponding slot is congested, the processing queue capacity of that slot can be appropriately increased, or some low-priority video data frames can be temporarily scheduled to other idle or less loaded slots. For data streams with extremely high timeliness requirements, such as target data, when the corresponding slot is congested, priority can be given to ensuring its processing resources, and the queue space of other non-critical data streams can be compressed.

[0049] Clearly, by monitoring slot load in real time and dynamically adjusting the allocation strategy according to data flow characteristics, slot congestion can be effectively alleviated, avoiding a decrease in the overall system processing efficiency due to local congestion. At the same time, it meets the differentiated processing needs of different types of data flows, further optimizing the overall system performance and improving the system's stability and real-time performance.

[0050] Based on the above method embodiments, this application also provides a dynamic slot management device for multi-source heterogeneous data streams of swarmed UAVs, comprising: an identifier allocation module, used to receive multi-source heterogeneous data streams of swarmed UAVs and assign a unique stream identifier to each data stream, wherein the data streams are classified according to data type, the data types including: video data, SAR radar data, and target data; a data distribution module, used to search for existing data processing slots to see if there are data processing slots bound to the same data stream identifier, and if so, directly use the data processing slot; if not, and there are free data processing slots, allocate a free data processing slot to the data stream; if all data processing slots are occupied, execute a timestamp-based LRU eviction policy, and finally distribute the data frame to the corresponding data processing slot, wherein the data processing slot is bound one-to-one with the data stream identifier, the most recent access timestamp, and the corresponding processing worker thread; or dynamically adjust the data processing slot allocation policy based on the data stream characteristics; and an update module, used to update the most recent access timestamp of the data processing slot to the current system time.

[0051] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A dynamic slot management method for multi-source heterogeneous data streams from swarmed UAVs, characterized in that, include: The system receives multi-source heterogeneous data streams from a cluster of UAVs and assigns a unique stream identifier to each data stream. The data streams are classified according to their data types, which include video data, SAR radar data, and target data. The system checks if any existing data processing slots exist and are bound to the same data stream identifier. If they do, the data processing slot is used directly. If not, and there are available data processing slots, an available data processing slot is allocated to the data stream. If all data processing slots are occupied, a timestamp-based LRU eviction policy is executed, and the data frame is finally distributed to the corresponding data processing slot. Here, each data processing slot is bound to a data stream identifier, the most recently accessed timestamp, and the corresponding processing worker thread in a one-to-one correspondence. Alternatively, the data processing slot allocation strategy can be dynamically adjusted based on the characteristics of the data stream. Update the most recently accessed timestamp of this data processing slot to the current system time.

2. The dynamic slot management method for multi-source heterogeneous data streams of clustered UAVs as described in claim 1, characterized in that, The execution of the timestamp-based LRU eviction policy includes: Initialize the LRU index and minimum timestamp; Iterate through all data processing slots and compare the most recent access timestamps of each data processing slot. Determine the data processing slot with the smallest timestamp as the data processing slot to be eliminated. Rebind the data processing slot to be phased out to a new stream identifier and update its most recently accessed timestamp to the current timestamp.

3. The dynamic slot management method for multi-source heterogeneous data streams of clustered UAVs as described in claim 1, characterized in that, Before receiving the multi-source heterogeneous data stream from the cluster of drones, the method further includes: Initialize the corresponding slot pool for each data stream distributor to obtain a slot pool where all data processing slots are idle; The number of data processing slots in the slot pool is fixed and determined based on the data processing capacity. The type of the slot pool corresponds one-to-one with the data type.

4. The dynamic slot management method for multi-source heterogeneous data streams of clustered UAVs as described in claim 2, characterized in that, The step of checking whether there is a data processing slot with the same data flow identifier among the existing data processing slots includes: The new data stream is assigned to the corresponding slot pool based on its unique identifier. Then, search the slot pool for a data processing slot that is bound to the same flow identifier.

5. The dynamic slot management method for multi-source heterogeneous data streams of clustered UAVs as described in claim 2, characterized in that, The step of traversing all data processing slots and comparing the most recent access timestamps of each data processing slot to determine the data processing slot with the smallest timestamp as the data processing slot to be phased out includes: Compare the timestamp of the current data processing slot with the smallest timestamp that has been recorded; If the timestamp of the current data processing slot is smaller, update the minimum timestamp to the timestamp of the current data processing slot and record the index of that slot; After the traversal is complete, the slot corresponding to the recorded slot index is used as the elimination target.

6. The dynamic slot management method for multi-source heterogeneous data streams of clustered UAVs as described in claim 1, characterized in that, After updating the most recently accessed timestamp of the data processing slot to the current system time, the method further includes: Data frames from multi-source heterogeneous data streams are processed, including video decoding, image decoding, and target parameter value parsing. Update the most recently accessed timestamp to the current system timestamp.

7. The dynamic slot management method for multi-source heterogeneous data streams of clustered UAVs as described in claim 1, characterized in that, After allocating an idle data processing slot to the data stream, the method further includes: Synchronize and update the timestamps of the data processing slots.

8. The dynamic slot management method for multi-source heterogeneous data streams of clustered UAVs as described in claim 1, characterized in that, The method further includes: Regularly check the usage status of data processing slots, and automatically unbind and mark data processing slots that have not received data for longer than the preset timeout threshold as idle.

9. The dynamic slot management method for multi-source heterogeneous data streams of clustered UAVs as described in claim 1, characterized in that, Based on a data flow characteristic-based strategy for dynamically adjusting data processing slot allocation, including: Monitor the processing queue length and processing latency of each data processing slot; When data processing congestion is detected, the queue size is dynamically adjusted according to the data stream characteristics, and the allocation of data processing slots in the data stream is dynamically adjusted according to the queue size.

10. A dynamic slot management device for multi-source heterogeneous data streams from swarmed UAVs, characterized in that, include: The identifier allocation module is used to receive multi-source heterogeneous data streams from a cluster of UAVs and assign a unique stream identifier to each data stream. The data streams are classified according to their data types, which include video data, SAR radar data, and target data. The data distribution module checks if any existing data processing slots exist and are bound to the same data stream identifier. If they do, the data processing slot is used directly. If not, and there are available data processing slots, an available data processing slot is allocated to the data stream. If all data processing slots are occupied, a timestamp-based LRU eviction policy is executed to distribute the data frame to the corresponding data processing slot. Each data processing slot is bound to a data stream identifier, a recent access timestamp, and a corresponding processing worker thread. Alternatively, the data processing slot allocation strategy can be dynamically adjusted based on the data stream characteristics. The update module is used to update the most recently accessed timestamp of the data processing slot to the current system time.

Images