Obstacle start-stop state detection method and system for autonomous driving

Abstract

The present invention discloses a method and system for detecting the start-stop state of an obstacle for autonomous driving, and relates to the technical field of autonomous driving. The method of the present invention can improve the accuracy of obstacle start-stop state machine detection and the robustness of decision-making by establishing obstacle information in an obstacle dictionary and detecting the start-stop state of the obstacle through the start-stop information, thereby improving the safety of the autonomous driving system.

Description

Technical Field

[0001] The present invention discloses a method and system for detecting the start-stop state of an obstacle for autonomous driving, and relates to the technical field of autonomous driving. Background Art

[0002] The detection of the start and stop status of obstacles in autonomous driving plays a vital role in decision-making. Generally, the obstacle start and stop status detection method is mainly based on the output of the perception module. However, there are certain errors in the output of the perception module. It is difficult to accurately estimate the speed and position of the obstacle in every frame at low speeds, resulting in low accuracy in the detection of the start and stop status of obstacles. It is impossible to make a good judgment on the start and stop intentions of obstacles such as vehicles and bicycles. In other words, the decision made solely based on the perception results is unstable, which will make the planning made by the planning module easy to jump, which has a bad impact in many scenarios of autonomous driving (such as deciding whether to bypass or stop in the queuing process). Summary of the invention

[0003] In view of the problems existing in the prior art, the present invention provides a method and system for detecting the start-stop state of an obstacle for autonomous driving. The technical solution adopted is:

[0004] In a first aspect, a method for detecting the start-stop state of an obstacle for autonomous driving is provided, the method comprising:

[0005] According to the customized application scenario, the obstacles to be detected are screened, including:

[0006] By initializing the obstacle dictionary, the obstacle information of the start-stop state is stored;

[0007] According to the obstacle information to be detected, the obstacle information is added to the obstacle dictionary through the obstacle statistics tool;

[0008] According to the screened obstacles, the start / stop status of the obstacles is detected by establishing a start / stop information dictionary, which specifically includes:

[0009] Establish a start-stop information dictionary and retain the information of the previous frame to the current frame;

[0010] Traversing the obstacle dictionary and initializing the defined start and stop information;

[0011] Traversing each element of the defined start-stop information, assigning all the defined start-stop information to the hash table in the start-stop information query device, and performing obstacle start-stop state detection;

[0012] The obstacle dictionary is traversed to initialize the defined start and stop information, specifically including:

[0013] Mark the object defining the start and stop information, update the obstacle information being traversed into the obstacle object, and update the current timestamp;

[0014] Initialize the obstacle decision type object and store the obstacle decision type of the previous frame;

[0015] According to the start-stop information dictionary, the id of the obstacle currently traversed is searched;

[0016] According to the core update function UpdateStartStopInfo(), the queue defining the start and stop information is updated;

[0017] Add the updated object defining the start / stop information to the start / stop information dictionary, and adjust the length of the start / stop information definition queue according to the obstacle ID;

[0018] Among them, each element of the defined start and stop information is traversed, all the defined start and stop information is assigned to the hash table in the start and stop information query device, and the obstacle start and stop status detection is performed, specifically including, according to the defined start and stop information queue of each element, comparing the timestamp of the last element with the current timestamp, and deleting the defined start and stop information that is greater than the custom time interval.

[0019] In some implementations, according to the core update function UpdateStartStopInfo(), the queue defining the start and stop information is updated, specifically including:

[0020] Read the obstacle state queue in the obstacle object and record the length of the queue as the history length;

[0021] Constructing a two-dimensional matrix of historical position information according to the historical length;

[0022] The number of elements of the vector is constructed according to the history length, and the history speed information in the obstacle state queue is stored.

[0023] In a second aspect, a system for detecting the start and stop state of an obstacle for autonomous driving is provided, the system comprising:

[0024] The obstacle screening module is used to screen obstacles that need to be detected according to custom application scenarios, including:

[0025] An information storage unit, used to store obstacle information in a start-stop state by initializing an obstacle dictionary;

[0026] A dictionary optimization unit, used for adding obstacle information to be detected according to the need to be detected into the obstacle dictionary through an obstacle statistics tool;

[0027] The state detection module is used to detect the start and stop state of the obstacle by establishing a start and stop information dictionary according to the screened obstacles, specifically including:

[0028] An information processing unit, used to establish a start-stop information dictionary and retain the information of the previous frame to the current frame;

[0029] An initial processing unit, used to traverse the obstacle dictionary and perform initialization processing on the defined start and stop information;

[0030] An obstacle analysis unit, used for traversing each element of the defined start-stop information, assigning all the defined start-stop information to a hash table in the start-stop information query device, and performing obstacle start-stop state detection;

[0031] Wherein, the initial processing unit specifically includes:

[0032] A marking update subunit, used for marking the object defining the start and stop information, updating the obstacle information being traversed into the obstacle object, and updating the current timestamp;

[0033] A decision processing subunit, used to initialize the obstacle decision type object and store the obstacle decision type of the previous frame;

[0034] An ID search subunit, used to search for the ID of the obstacle currently traversed according to the start-stop information dictionary;

[0035] A queue update subunit, used to update the queue defining the start and stop information according to a core update function UpdateStartStopInfo();

[0036] A queue adjustment subunit, configured to add the updated object defining the start / stop information to the start / stop information dictionary, and adjust the length of the start / stop information definition queue according to the obstacle ID;

[0037] The obstacle analysis unit is used to compare the timestamp of the last element with the current timestamp according to the defined start and stop information queue of each element, and delete the defined start and stop information that is greater than the custom time interval.

[0038] In some implementations, the queue updating subunit specifically includes:

[0039] The queue processing component is used to read the obstacle status queue in the obstacle object and record the length of the queue as the historical length;

[0040] A location information component, used to construct a two-dimensional matrix of historical location information according to the historical length;

[0041] The vector storage component is used to construct the number of elements of the vector according to the historical length, and store the historical speed information in the obstacle state queue.

[0042] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein when the one or more computer instructions are executed by the processor, the method described in the first aspect above is implemented.

[0043] In a fourth aspect, an embodiment of the present invention provides a computer storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it is used to implement the method described in the first aspect.

[0044] Compared with the prior art, the present invention has the following beneficial effects: the method of the present invention establishes obstacle information by establishing an obstacle dictionary, and detects the start and stop status of the obstacle through the start and stop information, thereby improving the accuracy of obstacle start and stop state machine detection and improving the robustness of decision-making, thereby achieving an improvement in the safety of the autonomous driving system. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0046] Figure 1 It is a flow chart of a method for detecting the start-stop state of an obstacle for autonomous driving provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] It is worth noting that the methods used in the present invention are all conventional methods unless otherwise specified; the raw materials and devices used are all conventional commercially available products, and their sources are not specifically limited unless otherwise specified.

[0048] Embodiment 1:

[0049] Figure 1 A flow chart of a method for detecting the start-stop state of an obstacle for autonomous driving is shown, Figure 1 As shown, the obstacle start-stop state detection method for autonomous driving provided in this embodiment includes:

[0050] S1, according to the custom application scenario, screen the obstacles that need to be detected, including:

[0051] S11, storing obstacle information of the start-stop state by initializing and setting an obstacle dictionary;

[0052] S12, adding obstacle information to be detected to the obstacle dictionary through an obstacle statistics tool;

[0053] S2, based on the selected obstacles, detecting the start / stop status of the obstacles by establishing a start / stop information dictionary, specifically including:

[0054] S21, establishing a start-stop information dictionary, and retaining the information of the previous frame to the current frame;

[0055] S22, traversing the obstacle dictionary and initializing the defined start and stop information;

[0056] S23, traversing each element of the defined start-stop information, assigning all the defined start-stop information to the hash table in the start-stop information query device, and performing obstacle start-stop state detection.

[0057] First, according to S1, according to the custom application scenario, the obstacles that need to be detected are screened out; for example, in the scenario of changing lanes or borrowing lanes, the obstacles within a certain distance before and after the current and target lanes are focused on; in the scenario of turning around, parking, getting out of trouble, etc., the obstacles within a certain area around the vehicle are focused on; in the scenario of driving or queuing, the obstacles within a certain distance before and after the current lane are focused on; in this embodiment, the upstream a priori provides a traffic flow statistics device, which counts all the obstacle information that needs to be considered in the current lane, the left lane, and the right lane. The vehicle can know the obstacle information that needs to be considered in the current lane, the left lane, and the right lane, and then screen out the obstacles that need to be detected according to the application scenario;

[0058] Specifically, according to S11, initialize an obstacle dictionary all_obs (type is map<int,ObstacleShrPtr> ), the "key-value" is "obstacle id-obstacle object", which is used to store the obstacle information that needs to be detected for start-stop status; customize the obstacle statistics tool GetAllObs() to traverse all obstacle information provided by the traffic flow statistics tool; if the traversed obstacle id is in all_obs, skip the obstacle; if the traversed obstacle type is the obstacle type to be considered (customized according to the business, such as vehicles and bicycles need to be considered, but fences do not), add the obstacle id and obstacle object to all_obs.

[0059] According to S12, the obstacle statistics tool GetAllObs() is used to add the obstacle information that needs to be detected among the obstacles provided by the traffic flow statistics module to all_obs.

[0060] Next, according to S2, specifically, according to S21, a static start-stop information dictionary map_int_deque is established, which is similar to the hash table in the start-stop information query StartStopInfoMap, that is, its type is also map <int, deque <startstopinfo>>, in addition, the meaning of static is: the dictionary will not be destroyed when each frame is run, and the information of the dictionary in the previous frame will be retained in the current frame;

[0061] Then, according to S22, the obstacle dictionary is traversed to initialize the defined start and stop information; specifically, the process includes:

[0062] According to S221, a StartStopInfo object is initialized and recorded as info, and the obstacle information being traversed is updated to its obs_ptr, and the current timestamp is updated to its getestamp;

[0063] Next, according to S222, the decision type object is initialized to last_type, which is used to store the obstacle decision type of the previous frame;

[0064] Next, S223, the id of the obstacle currently traversed is input into map_int_deque for search. If the StartStopInfo queue of the obstacle is found, the decision type of the last element of the queue (i.e., the most recently stored result) is assigned to last_type; if the StartStopInfo queue of the obstacle is not found, last_type is set to NO_IDEA.

[0065] Next, according to S224, enter the core update function UpdateStartStopInfo(); This function is a core update function, used to update the StartStopInfo object; This function passes in info and last_type, and this function passes out the updated info;

[0066] Next, according to S225, the updated info is added to the StartStopInfo queue corresponding to the obstacle id currently traversed in map_int_deque. If the queue length exceeds the preset value (custom length), the earliest stored element is deleted;

[0067] Then, according to S23, all elements in map_int_deque are traversed; the timestamp of the last element (i.e., the most recently stored result) of the StartStopInfo queue of each element (i.e., each obstacle) is compared with the current timestamp. If it is greater than the preset value (custom time interval), the StartStopInfo queue corresponding to the obstacle is deleted; this step ensures that map_int_deque only focuses on the most recently appeared obstacles, and the obstacle information that has not been updated for a period of time is deleted; all map_int_deque values ​​of this frame are assigned to the hash table in the global start-stop information query StartStopInfoMap for subsequent module queries.

[0068] Embodiment 2:

[0069] On the basis of the first embodiment, S224 specifically includes:

[0070] S2241, reading the obstacle state queue in the obstacle object, and recording the length of the queue as the history length;

[0071] S2242, constructing a two-dimensional matrix of historical position information according to the historical length;

[0072] S2243: construct the number of elements of the vector according to the history length, and store the history speed information in the obstacle state queue.

[0073] Read the obstacle state queue ObstacleStateHistory in the obstacle object obs_ptr in info, and record the length of the queue as history_size;

[0074] Construct a matrix position_vec with history_size rows and 2 columns to store the historical position information in the obstacle state queue, where the first column is the historical position x and the second column is the historical position y;

[0075] Construct a vector velocity_vec with history_size elements to store the historical velocity information in the obstacle state queue.

[0076] Design a decision execution function DecisionExecution() and execute it. The following is the function content:

[0077] - Calculate the mean of vector velocity_vec and assign it to mean_velocity in info;

[0078] - Calculate the variance of vector position_vec through the function CalcVariance() and assign it to var_position in info

[0079] - Determine the number of predicted lines in the obstacle object obs_ptr;

[0080] - If the number of predicted lines is greater than 0, the obstacle object is judged to be started, the type of info is set to START, the judgment ends, and the function exits;

[0081] - If the number of predicted lines is equal to 0, it is judged that the obstacle object may be stopped, and the type of info is not processed first, and the next judgment is carried out;

[0082] - Determine whether the number of history_size is less than the custom threshold less_history_size;

[0083] - If it is less than, it is judged that the obstacle has just appeared and the corresponding data confidence is not high enough. At this time, the following judgment is made;

[0084] - If mean_velocity in info is greater than the custom threshold, it is judged that the obstacle object is likely to be started, and the type of info is set to START_LIKE, ending the judgment and exiting the function;

[0085] - If mean_velocity in info is less than the custom threshold, it is judged that the obstacle object is likely to be stopped, and the type of info is set to STOP_LIKE, ending the judgment and exiting the function;

[0086] - If it is greater than, it is judged that the obstacle has existed for a period of time, and the corresponding data confidence is high. The type of info is not processed first, and the next judgment is made;

[0087] - Determine whether mean_velocity is greater than the custom threshold mean_velocity_threshold;

[0088] - If it is greater than, the obstacle object is judged to be started, the type of info is set to START, the judgment ends, and the function exits;

[0089] - If it is less than, it is judged that the obstacle object may be stopped, and the type of info is not processed first, and the next judgment is carried out;

[0090] - Determine whether var_position is less than the custom threshold var_position_threshold;

[0091] - If it is less than, the obstacle object is judged to be stopped, the type of info is set to STOP, the judgment ends, and the function exits;

[0092] - If it is greater than, it is determined that the obstacle object may be activated, and the type of info is not processed first, and the next judgment is made;

[0093] - First, use the function CalcRSquared() to calculate the correlation coefficient of the vector position_vec, assign it to r_squared in info, and then determine the value of r_squared:

[0094] - If r_squared is greater than or equal to 0.9, the obstacle object is judged to be started, the type of info is set to START, the judgment ends, and the function exits;

[0095] - If r_squared is greater than or equal to 0.5 but less than 0.9, it is judged that the obstacle object is most likely started, and the type of info is set to START_LIKE, the judgment ends, and the function exits;

[0096] - If r_squared is greater than or equal to 0.1 but less than 0.5, it is judged that the obstacle object is likely to be stopped, and the type of info is set to STOP_LIKE, ending the judgment and exiting the function;

[0097] - If r_squared is less than 0.1, the obstacle object is judged to be stopped, the type of info is set to STOP, the judgment ends, and the function exits;

[0098] Implementation of the function CalcVariance():

[0099] - Calculate the mean of the vector position_vec, denoted as mean_position;

[0100] - Calculate the difference between vector position_vec and mean_position, recorded as diff_vec;

[0101] - Calculate the square of the vector diff_vec, denoted as square_vec;

[0102] - Calculate the mean of vector square_vec, denoted as mean_square;

[0103] - Calculate the square root of mean_square and assign it to var_position in info;

[0104] Implementation of the function CalcRSquared():

[0105] - Construct a design matrix X, where the number of rows is history_size, the number of columns is 2, the first column is the first column of the vector position_vec, that is, the historical sequence of position x, and the second column is a vector of all 1s;

[0106] - Construct the dependent variable vector Y, where the number of elements is history_size, and the element value is the second column of the vector position_vec, that is, the historical sequence of position y;

[0107] - Use linear least squares to solve the regression coefficients coeffs = (X.transpose() * X).inverse()* X.transpose() * Y;

[0108] - Calculate the total sum of squares:

[0109] - Calculate the mean of Y, denoted as mean_Y;

[0110] - Calculate the square of the difference between each element of Y and mean_Y, and then add up these square values, recorded as ss_total

[0111] - Calculate the residual sum of squares:;

[0112] - Calculate the product of the dependent variable vector Y and the regression coefficient vector coeffs, recorded as Y_pred;

[0113] - Calculate the square of the difference between each element of Y_pred and Y, and then add up these square values, recorded as ss_residual;

[0114] - Calculate the correlation coefficient R2 = 1 - ss_residual / ss_total and assign it to r_squared in info.

[0115] Embodiment three:

[0116] An embodiment of the present invention provides an obstacle start-stop state detection system for autonomous driving, the system comprising:

[0117] The obstacle screening module is used to screen obstacles that need to be detected according to custom application scenarios, including:

[0118] An information storage unit, used to store obstacle information in a start-stop state by initializing an obstacle dictionary;

[0119] A dictionary optimization unit, used for adding obstacle information to be detected according to the need to be detected into the obstacle dictionary through an obstacle statistics tool;

[0120] The state detection module is used to detect the start and stop state of the obstacle by establishing a start and stop information dictionary according to the screened obstacles, specifically including:

[0121] An information processing unit, used to establish a start-stop information dictionary and retain the information of the previous frame to the current frame;

[0122] An initial processing unit, used to traverse the obstacle dictionary and perform initialization processing on the defined start and stop information;

[0123] The obstacle analysis unit is used to traverse each element of the defined start-stop information, assign all the defined start-stop information to the hash table in the start-stop information query device, and perform obstacle start-stop state detection.

[0124] Specifically, the initial processing unit specifically includes:

[0125] A marking update subunit, used for marking the object defining the start and stop information, updating the obstacle information being traversed into the obstacle object, and updating the current timestamp;

[0126] A decision processing subunit, used to initialize the obstacle decision type object and store the obstacle decision type of the previous frame;

[0127] An ID search subunit, used to search for the ID of the obstacle currently traversed according to the start-stop information dictionary;

[0128] A queue updating subunit, used for updating the queue defining the start and stop information according to a core update function;

[0129] The queue adjustment subunit is used to add the updated object defining the start / stop information to the start / stop information dictionary, and adjust the length of the start / stop information definition queue according to the obstacle ID.

[0130] Specifically, the obstacle analysis unit is used to compare the timestamp of the last element with the current timestamp according to the defined start and stop information queue of each element, and delete the defined start and stop information that is greater than the custom time interval.

[0131] Specifically, the queue updating subunit includes:

[0132] The queue processing component is used to read the obstacle status queue in the obstacle object and record the length of the queue as the historical length;

[0133] A location information component, used to construct a two-dimensional matrix of historical location information according to the historical length;

[0134] The vector storage component is used to construct the number of elements of the vector according to the historical length, and store the historical speed information in the obstacle state queue.

[0135] Embodiment 4:

[0136] This embodiment further provides an electronic device, including a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions implement the method of embodiment 1 when executed by the processor;

[0137] In practical applications, the processor can be an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a controller, a microcontroller unit (MCU), a microprocessor or other electronic components to execute the methods in the above embodiments.

[0138] The method implemented in this embodiment is as described in Embodiment 2.

[0139] Embodiment five:

[0140] This embodiment further provides a computer storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by one or more processors, the method of the first embodiment is implemented;

[0141] Among them, the computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0142] The method implemented in this embodiment is as described in Embodiment 2.

[0143] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions of the technical solution of the present invention by ordinary technicians in this field do not deviate from the essence and scope of the technical solution of the present invention.< / startstopinfo>

Claims

1. A method for detecting the start-stop state of an obstacle for autonomous driving, characterized in that: The method comprises: According to the customized application scenario, the obstacles to be detected are screened, including: By initializing the obstacle dictionary, the obstacle information of the start-stop state is stored; According to the obstacle information to be detected, the obstacle information is added to the obstacle dictionary through the obstacle statistics tool; According to the screened obstacles, the start-stop status of the obstacles is detected by establishing a start-stop information dictionary, which specifically includes: Establish a start-stop information dictionary and retain the information of the previous frame to the current frame; Traversing the obstacle dictionary and initializing the defined start and stop information; Traversing each element of the defined start-stop information, assigning all the defined start-stop information to the hash table in the start-stop information query device, and performing obstacle start-stop state detection; The obstacle dictionary is traversed to initialize the defined start and stop information, specifically including: Mark the object defining the start and stop information, update the obstacle information being traversed into the obstacle object, and update the current timestamp; Initialize the obstacle decision type object and store the obstacle decision type of the previous frame; According to the start-stop information dictionary, the id of the obstacle currently traversed is searched; According to the core update function, the queue defining the start and stop information is updated; Add the updated object defining the start / stop information to the start / stop information dictionary, and adjust the length of the start / stop information definition queue according to the obstacle ID; Among them, traversing each element of the defined start and stop information, assigning all the defined start and stop information to the hash table in the start and stop information query device, and performing obstacle start and stop state detection, specifically including, according to the defined start and stop information queue of each element, comparing the timestamp of the last element with the current timestamp, and deleting the defined start and stop information that is greater than the custom time interval; Wherein, according to the core update function, the queue defining the start and stop information is updated, specifically including: Read the obstacle state queue in the obstacle object and record the length of the queue as the history length; Constructing a two-dimensional matrix of historical position information according to the historical length; The number of elements of the vector is constructed according to the history length, and the history speed information in the obstacle state queue is stored.

2. An obstacle start-stop state detection system for autonomous driving, characterized in that: The system comprises: The obstacle screening module is used to screen obstacles that need to be detected according to custom application scenarios, including: An information storage unit, used to store obstacle information in a start-stop state by initializing an obstacle dictionary; A dictionary optimization unit, used for adding obstacle information to be detected according to the need to be detected into the obstacle dictionary through an obstacle statistics tool; The state detection module is used to detect the start and stop state of the obstacle by establishing a start and stop information dictionary according to the screened obstacles, specifically including: An information processing unit, used to establish a start-stop information dictionary and retain the information of the previous frame to the current frame; An initial processing unit, used to traverse the obstacle dictionary and perform initialization processing on the defined start and stop information; An obstacle analysis unit, used for traversing each element of the defined start-stop information, assigning all the defined start-stop information to a hash table in the start-stop information query device, and performing obstacle start-stop state detection; Wherein, the initial processing unit specifically includes: A marking update subunit, used for marking the object defining the start and stop information, updating the obstacle information being traversed into the obstacle object, and updating the current timestamp; A decision processing subunit, used to initialize the obstacle decision type object and store the obstacle decision type of the previous frame; An ID search subunit, used to search for the ID of the obstacle currently traversed according to the start-stop information dictionary; A queue updating subunit, used for updating the queue defining the start and stop information according to a core update function; A queue adjustment subunit, configured to add the updated object defining the start / stop information to the start / stop information dictionary, and adjust the length of the start / stop information definition queue according to the obstacle ID; The obstacle analysis unit is used to compare the timestamp of the last element with the current timestamp according to the defined start and stop information queue of each element, and delete the defined start and stop information that is greater than the custom time interval; Wherein, the queue updating subunit specifically includes: The queue processing component is used to read the obstacle status queue in the obstacle object and record the length of the queue as the historical length; A location information component, used to construct a two-dimensional matrix of historical location information according to the historical length; The vector storage component is used to construct the number of elements of the vector according to the historical length, and store the historical speed information in the obstacle state queue.

3. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions implement the method as claimed in claim 1 when executed by the processor.

4. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it is used to implement the method as claimed in claim 1.

Images