JP6548708B2 - Low Latency Testing Machine for Image Processing Systems - Google Patents

Description

  The present invention relates to a tester for an image processing system, in particular a support system for performing image processing or a tester for automatic control of a vehicle. Transport should be understood as any device configured to move on its own, such as a land vehicle, an aircraft, a boat or a submarine.

  For many years, hardware-in-the-loop simulation has been established as part of a series of development and evaluation of safety critical electronic controls. In this case, a prototype of the control device is connected to the simulator, which simulates the environment of the control device based on the software, wherein for example the simulation of the sensor creates data for data entry of the control device. To the data input unit. Conversely, the simulator reads the data from the data output of the controller and takes these data into account, for example, in the calculation of the next time step of the simulation by the simulation of the actuator. Such a simulator can also be configured as a test machine, in which case, in addition to the control device, other physical components which are likewise integrated in the simulation, which cooperate with this control device, the motor vehicle In the case of a control device, for example, a steering system, a motor or an imaging sensor can be included. The control device thus operates in a fully virtual environment, in which it can be reproducibly tested without risk in several different situations.

  The controller operates in hard real time because it is provided to control, coordinate or monitor the physical system. Therefore, the simulator also has to operate in hard real time, ie within a defined time interval, for example 1 ms, it must be ensured that the calculation of all data requested by the controller is complete.

  In recent years, the automotive industry has developed a number of driver assistance systems, which use imaging sensors with different technologies to image around the vehicle, such as radar images, lidar images, or Lens-based optical images are created, these images are read by the control device for evaluation and interpretation, and in the case of experimental autonomous vehicles either to intervene in driving behavior based on the read-in images, On the contrary, it independently controls the vehicle independently of the human driver. Examples are radar based adaptive cruise control, pedestrian recognition or traffic sign recognition.

  Therefore, the tester for such a support system must be configured to calculate the expected image by the controller and supply it to the controller. In this case, there is a problem that the amount of calculation for calculating the image is very large, and it takes time accordingly. According to the state of the art, it takes well 50 to 100 ms to calculate a two-dimensional projection as detected by an imaging sensor from a three-dimensional ambient model as stored in simulation software there is a possibility. Such high latency is not compatible with the above-mentioned requirements for the real-time property of the test machine, and makes the simulation results unconvincing.

  German Utility Model 202015104345 (DE 20 2015 104 345 U1) describes a test machine for a control device which carries out image processing, which test machine is for the control device by means of an adapter module. To reduce the latency of the image data, the adapter module bypasses the imaging sensor and supplies the image data directly to the control device to provide a shortened data path for the image data. However, the latency generated by the calculation of the image data can not be compensated for in this way.

  With the above background in mind, it is an object of the present invention to reduce the latency inaccuracies that arise due to the calculation of image data for an image processing system in a tester. The above problem is solved by a tester having the features of claim 1 or a method having the features of claim 15. Advantageous embodiments are the subject of the dependent claims.

  The present invention describes, at its core, a method for at least partially compensating latency by time extrapolation of an ambient model stored in the simulator based on the measurement of latency. The invention is in particular a tester having a first arithmetic unit, in particular a processor (CPU), which is programmed by simulation software with respect to a surrounding model. The simulation software periodically calculates the first position and the first velocity vector, preferably in hard real time, for the first virtual object in the surrounding model, for example for the virtual vehicle, It is at least configured to correspond to a virtual object of The second computing unit of the testing machine, which preferably comprises at least a graphics processor (GPU), comprises first image data, in particular of a first virtual object, representing a two-dimensional first projection of the surrounding model. It is configured to calculate first image data that reproduces an image of the imaging sensor. To this end, the second arithmetic unit is configured to periodically read the first position of the first virtual object and to calculate first image data based on this position.

  The tester further includes an adapter module for incorporating the image processing system into the tester or simulation. The adapter module reads the first image data, emulates a first imaging sensor of the image processing system to process the first image data, and supplies the processed first image data to the image processing system Is configured as.

  Thus, that is, if the image processing system processes an image of an optical lens-based camera as an example, the adapter module records and processes the first image data, in this case the processed first image. The data correspond to the data that the camera's optical sensor will supply to the imaging system in the situation reproduced by the simulation software. That is, the adapter module acts as a so-called substitute for the imaging sensor of the imaging system during simulation, and emulates the imaging sensor and supplies expected imaging data to the imaging system instead of the imaging sensor.

  Further, the first arithmetic unit reads control data calculated for the actuator by the image processing system based on the processed first image data, and takes into account the control data, the first virtual unit It is configured to associate the object with a new velocity vector. That is, for example, when the image processing system is a driving support system and outputs control data for starting an automatic braking maneuver based on the identified dangerous situation, the first arithmetic unit is a surrounding model The first virtual object in, in this case a virtual vehicle, is configured to imitate.

  In order to compensate for the latency generated by the calculation of the first image data, the tester completes the processing of the first image data by the adapter module from the start of the calculation of the first image data by the second arithmetic unit. It is configured to determine the length Δt of the time interval that elapses up to. The measured length Δt is stored at the memory address, read by the first arithmetic unit, and used to estimate the latency L of the first image data. The estimated latency L takes into account the first position x (t) of the first virtual object, the first velocity vector v (t) of the first virtual object, and the estimated latency L Then, it is used by the first operation unit or simulation software running on the first operation unit to obtain the first extrapolated position x (t + L).

  Thus, the first extrapolated position x (t + L) is an estimate of the future position of the first virtual object at time t + L, where t is the current time in system time or simulation. (This is equivalent because the simulation is performed in hard real time.) Here, the second arithmetic unit reads the current first position x (t) to calculate the first image data Rather, it is configured to read the first extrapolated position x (t + L). That is, the latency of the first image data is at least partially compensated by assuming that the second arithmetic unit assumes the future state of the simulated surrounding model from the beginning when calculating the first image data. Ru. When the first image data thus calculated is finally supplied to the image processing system, in an ideal case, the surrounding model on the first arithmetic unit also reaches this future state, This will cause the first image data in the image processing system to match the current state of the surrounding model, and the tester will supply real data.

  In principle, any numerical integration method can be used to determine the first extrapolated position, for example the first or higher order Runge-Kutta method. The present invention does not guarantee that the inaccuracies caused by the latency of the first image data are completely compensated. In order to estimate the first extrapolation position, it is preferably integrated over the estimated latency L, the length of L being usually much longer than the time step in the simulation of the surrounding model, so It can not be expected that the extrapolated position x (t + L) of 1 will coincide with the actual position that will be associated with the virtual object at time t + L. Furthermore, the estimated latency L may differ slightly from the actual latency of the first image data, since, for example, the amount of computation to calculate the first image data may vary depending on the state of the surrounding model there's a possibility that. However, according to the invention, at least an improvement of the simulation results is achieved, since the inaccuracies caused by the above-mentioned effects are smaller than the inaccuracies that the uncompensated first image data latency will cause. It is possible.

  Among other things, the first virtual object can be configured as a virtual vehicle, and the image processing system can be configured as an automatic control or assistance system for the vehicle.

  Preferably, the second arithmetic unit is configured to calculate the first image data such that the first projection mimics the field of view of the first imaging sensor. For this purpose, in the calculation of the first image data by the second arithmetic unit, the image processing system is the first virtual object image processing system, and the first imaging sensor is the first virtual It is implemented on the assumption that it is attached at a well-defined position of the object. In order to save computational complexity and thereby keep the latency of the first image data as low as possible from the beginning, the second arithmetic unit is based on the above assumption in order to calculate the first image data It is configured to consider only virtual objects of the surrounding model that are located inside the field of view of the first imaging sensor. In one possible embodiment, the image processing system is a radar based adaptive cruise control, for example for a motor vehicle, and thus the first imaging sensor is in this case the first virtual object in the surrounding model. Is assumed to be part of a radar system located on the front side of a virtual car. If the radar system is technically only configured to identify objects within the range of, for example, 200 m, then the 200 m range of the surrounding model and of the radar system may be calculated to calculate the first image data. Only virtual objects located within the field cone should be considered.

  Generally speaking, the field of view of the imaging sensor should be understood as the whole of all objects visible at a given time by the imaging sensor, in the form recognized by the imaging sensor, the second operation The unit is preferably arranged to initially consider only the information obtainable from the field of view of the first imaging sensor according to the above definition in order to calculate the first image data.

  In the case of the example described above, this also means, for example, that the first image data for the radar system should not contain information on the colors of virtual objects visible in the first projection, and Even if the above color information is present in the surrounding model, this means that this information is not taken into account from the outset when calculating the first image data.

  In one embodiment, the measurement of the time interval length Δt is a system in which the tester, in particular the second computing unit, reads out the system time of the tester at the start of the calculation of the first image data. It is carried out in such a manner that a time stamp in which time is stored is given to the first image data. After the adapter module processes the first image data, and before the adapter module supplies the first image data to the image processing system, the adapter module reads the time stamp and the system time stored in the time stamp The difference Δt between time intervals is determined by forming a difference between these two system times in comparison with the current system time, and the determined time interval length Δ is stored in a memory address accessible by the first operation module Do. The first arithmetic unit is configured to read the determined time interval length Δt at the memory address.

  In another embodiment, the time measurement is performed based on a digital ID that the tester, in particular the first computing unit or the second computing unit creates for the first image data. The digital ID is created prior to the calculation of the first image data and transmitted to the adapter module along with the first system time of the tester. In this case, the first system time is the system time at the time of transmission of the digital ID. After calculating the first image data, the second arithmetic unit assigns a digital ID to the second image data, and transmits the second image data together with the digital ID to the adapter module. The adapter module is configured to read the digital ID from the first image data and to associate the first image data based on the digital ID of the first system time. After completing the processing of the first image data, the adapter module compares the current system time of the tester with the first system time to determine the time interval length Δt, and sets this length Δt as the memory address. Remember.

  A prerequisite for both types of measurement is that the tester has a sufficiently fast synchronization of system time between the components of the tester.

  Advantageously, the components of the tester essential to the invention are each connected via a real-time capable data connection, which is configured for streaming data, ie It is configured to transmit a continuous stream of large amounts of data in real time. Therefore, specifically, a first data connection having real-time ability is formed between the first arithmetic unit and the second arithmetic unit, and real-time ability between the second arithmetic unit and the adapter module. A second data connection, preferably an HDMI connection, is formed, and a third data connection, preferably an Ethernet connection, having real-time capability is formed between the adapter module and the first computing unit.

  Particularly preferably, the first data connection is provided by a real time capable bus of the tester. This embodiment is advantageous in that it allows the second computing unit to be configured as an integral part of the tester, which has a favorable effect on latency. This is because the internal bus of a typical hardware in-the-loop simulator is optimized for real-time performance, ie low latency and low jitter.

  Further advantageously, the second computing unit is configured to operate at least a second imaging sensor as an option in addition to the first imaging sensor. For example, the image processing system can include a stereo camera, which causes the second computing unit to calculate two optical images, and the adapter module accordingly images the two optical images. It will have to be supplied to the processing system. In another example, the imaging system can include multiple controllers with multiple imaging sensors. Thus, in an advantageous embodiment, the second arithmetic unit is at least for the second imaging sensor of the image processing system in parallel with the calculation of the first image data or after the calculation of the first image data. The second image data representing a two-dimensional second projection of the surrounding model is calculated. Thereafter, the entire image data is preferably bundled and transmitted. For this purpose, the second arithmetic unit is configured to create data packets comprising the first image data and the second image data and, if present, further image data. If the measurement of the time interval Δt is performed using time stamps, the data packets are time stamped. The adapter module reads the data packet and, in addition to the processing of the first image data already described, also processes the second image data by emulation of the second imaging sensor and images the processed second image data It is configured to supply a processing system.

  Preferably, the estimated latency L is not a static parameter measured only once, and the first arithmetic unit dynamically adapts the estimated latency L value during the running simulation. It is configured to let you That is, the tester periodically obtains the length Δt of the time interval, and periodically reads it by the first arithmetic unit, and dynamically resets the value of L to the current latency of the first image data during the simulation. It is configured to fit.

  In a simple embodiment, this causes the first arithmetic unit to set the estimated latency L equal to the current value of the time interval length Δt, ie to L = Δt periodically. To be implemented. In another embodiment, the first computing unit stores a plurality of values of Δt measured in the past, and the latency L from the values of these plurality of Δt, in particular the average of the values of these plurality of Δt. It is configured to calculate as a value, a weighted average, or a median.

  The invention will be explained in more detail below on the basis of the drawings.

FIG. 1 is a very simplified schematic view of a tester known from the prior art for an image processing system. FIG. 1 is a schematic view of a tester according to the invention in a preferred embodiment.

  FIG. 1 is used to visualize the ideas underlying the present invention and the problems solved by the present invention. This figure outlines one test scenario using the tester represented by the simulation computer SIM and using the imaging system UUT as a subject. The image processing system UUT is, by way of example, a camera based accident support system configured to identify a dangerous situation in a vehicle and to initiate an automatic braking maneuver.

  A surrounding model MOD is stored in the simulator SIM. The ambient model MOD is software executable by the first processor C1 of the simulator SIM and configured to simulate the environment of the image processing system UUT and the test scenario for the image processing system UUT It is. The surrounding model MOD includes a plurality of virtual objects, and a subset of the virtual objects is movable. The movable virtual objects are associated with these virtual objects in addition to the position (vector value) in the surrounding model and one velocity vector respectively, and the virtual objects in the surrounding model MOD It is characterized in that the position of the object can be changed at each time step of the simulation. The illustrated surrounding model MOD includes, as an example, a first virtual vehicle VEH1 as a first virtual object, and a second virtual vehicle VEH2 as a second virtual object. The illustrated test scenario is an accident situation at an intersection. Two vehicles are virtual objects that can move. Therefore, the first virtual vehicle VEH1 is associated with the time-dependent first position x (t) and the time-dependent first velocity vector v (t), and the second virtual The time-dependent second position x ′ (t) and the time-dependent second velocity vector v ′ (t) are associated with the typical vehicle VEH2.

  Thus, the state of the surrounding model MOD at time t includes, as entries, the coordinates of the positions of all virtual objects and the entries of the velocity vectors of all movable virtual objects. It can be described by (t).

  The simulator SIM and the image processing system UUT construct a simulated control loop together. The simulator SIM continuously supplies emulated image data SE to the image processing system UUT, and the image processing system UUT uses the emulated image data SE as real image data, that is, a physical imaging sensor Interpret as image data supplied by Based on these image data, the image processing system UUT feeds back the control data AC to the simulator, and thus the simulator mimics the physical response of the vehicle to the control data AC in the first virtual vehicle VEH1. , Affect the state M of the surrounding model MOD.

  From when the calculation of the image data SE is started to when the image data SE is supplied to the image processing system UUT, a time interval of a length Δt which substantially occurs due to the calculation and processing of the image data SE elapses. Therefore, this length Δt corresponds to the latency of the image data. Assuming that the latency is Δt = 50 ms, this means that the exemplary accident support system in the simulation only recognizes the danger situation with a delay of 50 ms and responds accordingly. Such values would be unacceptable for the illustrated test scenario, and simulation results would only be able to be used in a limited way.

  The image data SE represents a two-dimensional projection of the field of view of the first imaging sensor, ie the surrounding model MOD, mounted at a predetermined position of the first virtual vehicle VEH1. In this regard, the image data SE should be understood as a function D [M (t)] of the state vector M (t). Therefore, the processed image data finally supplied to the image processing system UUT is defined by the function D [M (t-Δt)]. It is possible to basically compensate for latency by supplying future image data SE described by the function D [M (t + Δt)] from the simulator SIM to the image processing system UUT by substitution of t → t + Δt. Can be recognized directly. In that case, the image data supplied to the image processing system is described by the function D [M (t + Δt−Δt)] = D [M (t)], ie, the current state M (t of the surrounding model MOD (t). Match with).

  Basically, the future state M (t + Δt) of the surrounding model MOD is unknown. However, if the latency Δt is learned, eg by measurement, it is possible to at least estimate this future state M (t + Δt) by extrapolation of the current state M (t) over the length Δt, and the simulation accuracy It is possible to improve.

  FIG. 2 shows a schematic view of a tester configured for this. The testing machine comprises a host computer HST, a simulator SIM and an adapter module AD, the image processing system UUT a first control unit ECU1 for the radar system and a second control unit ECU2 for the stereo camera. And.

  The simulator SIM includes a first arithmetic unit CPU having a first processor C1, and the simulator SIM has a second arithmetic unit GPU having a second processor C2 and a graphics processor (graphics processing unit) C3. Including. The host computer HST is configured to store the surrounding model MOD in the first arithmetic unit CPU via the fifth data connection DL, and the first processor C1 is configured to execute the surrounding model ing. (The ambient model MOD illustrated in FIGS. 1 and 2 is assumed to be identical below.) The first computing unit CPU and the second computing unit GPU are the first with real-time capability of the tester. Connected together to the bus BS, this first bus BS thus provides a first data connection between the first arithmetic unit CPU and the second arithmetic unit GPU. The first bus BS is technically optimized for real-time performance and thus guarantees a low latency first data connection.

  The first computing unit CPU is configured to periodically transmit the position of the virtual object in the surrounding model to the second computing unit GPU via the first data connection BS. The second operation unit GPU reads the transmitted position, and uses the rendering software REN stored in the second operation unit GPU to generate the first image data, the second image data, and the third image data. Are configured to be calculated at least as a function of the transmitted position, in particular as a function of the first position x (t) and the second position x ′ (t).

  For this purpose, many shaders are implemented in the rendering software. The first shader calculates first image data. The first image data represents a first projection of the surrounding model MOD, which mimics the field of view of the radar sensor attached to the first virtual vehicle VEH1. The second shader calculates second image data and third image data. The second image data represents a second projection of the surrounding model, and the third image data represents a third projection of the surrounding model. The second and third projections mimic the field of view of the first and second photosensors of the camera optics mounted on the virtual vehicle VEH 1 respectively. For this purpose, the second shader is also configured to simulate, inter alia, the optics of the lens system of a stereo camera.

  At the same time as the transmission of the first location x (t) and the second location x '(t), the first arithmetic unit CPU transmits via the third data connection ETH with real-time capability configured as an Ethernet connection. The digital ID of the tester and the first system time are transmitted, and further the digital ID is transmitted to the second arithmetic unit GPU via the first data connection BS. The second arithmetic unit GPU creates a data packet including the first image data, the second image data, the third image data, and the digital ID. The graphics processor C3 transmits data packets to the adapter module AD via a second real-time capable data connection HDMI configured as an HDMI connection.

  The adapter module AD includes an FPGA F. Three parallel emulation logics are implemented on the FPGA F. The first emulation logic EM1 is configured to emulate the first imaging sensor of the radar system, ie to record the first image data, process this first image data and process it The first image data of (1) is configured to correspond to the image data expected by the first control unit ECU1. Correspondingly, the second emulation logic EM2 and the third emulation logic EM3 record the second image data or the third image data, and the second imaging sensor or the second of the lens-based optical stereo camera It is configured to emulate the three imaging sensors.

  The first controller from the adapter module AD is such that the processed first image data is interpreted by the first controller ECU1 as physical image data of a physical imaging sensor. It is supplied to the ECU 1. The technical means necessary for this are known in the art and can be easily understood by the person skilled in the art. Special development controllers often provide a dedicated interface for this.

  The first control unit ECU1 and the second control unit ECU2 control the transport means, in particular for the actuators of the vehicle, on the basis of the processed first image data or the processed second image data. Calculate the signal. The control signal is supplied to the first arithmetic unit CPU via the second bus XB, for example, the CAN bus, connected to the first bus BS via the gateway G and disposed outside the simulator SIM, It is read by the first processor C1 and in the calculation of the next time step of the simulation, it is taken into account that the physical vehicle response to the control signal in the first virtual vehicle VEH1 is reproduced.

  The adapter module AD is further configured to associate the data packet with the first system time acquired from the first arithmetic unit based on the digital ID. Specifically, this means that the adapter module AD reads the digital ID transmitted from the first arithmetic unit CPU via the third data connection ETH together with the first system time, and the adapter module further Reading out the digital ID stored in the data packet, comparing the read out two digital IDs to identify the same, and associating the first system time with the data packet based on the comparison means. Immediately after processing at least the first image data, the adapter module AD compares the first system time with the current system time of the tester and determines the length of the time interval Δt by difference formation, the value of this Δt Are transmitted to the first arithmetic unit CPU via the third data connection ETH. Preferably, this is not performed only once, and the adapter module AD periodically and continuously calculates the current value of Δt to calculate the current current value of Δt as the first operation. Transmit continuously to the unit CPU.

  In order to be able to perform the measurement, the adapter module needs to have access to the system time of the tester. In the illustrated embodiment, the adapter module is not connected to the first bus BS of the tester, so that, for example, the system time is continuously transmitted to the adapter module AD via the third data connection ETH. The adapter module AD synchronizes the local clock with the system time or, if necessary, directly reads the system time sent via the third data connection ETH.

  Basically, the digital ID can be omitted. In an alternative embodiment, the measurement of the length Δt is performed on the basis of a time stamp. The second computing unit GPU applies this timestamp to the data packet, and the second computing unit GPU uses this time for the first system time of the tester at a time before the first image data is calculated. Stored in the stamp, the adapter module reads the first system time from this timestamp.

  The first arithmetic unit CPU is configured to read the value of Δt and estimate the latency L of the first image data based on the value of Δt. In a simple embodiment this is implemented such that the first arithmetic unit CPU simply uses the current value of Δt from time to time for the estimated latency L. However, this embodiment can be problematic if short-term fluctuations occur in the latency of the image data. Advantageously, the first arithmetic unit CPU calculates the value of the estimated latency L based on a plurality of values of Δt measured in the past. For example, the first arithmetic unit CPU stores, for example, the latest 100 values of .DELTA.t from time to time, and the value of L is an average value, a weighted average value or a median value of the stored values of .DELTA.t. Can be configured to calculate.

  Here, in the illustrated embodiment, the first computing unit is to select for all movable virtual objects in the surrounding model, or at least to the associated movable virtual objects. Latency compensation is performed, in particular for calculating the extrapolated position using the estimated latency L, for the first virtual vehicle VEH1 and the second virtual vehicle VEH2 Ru. Therefore, the first arithmetic unit CPU generates a first extrapolation position x based on the first position x (t) and the first velocity vector v (t) for the first virtual vehicle VHE1. Calculate (t + L), and the first arithmetic unit CPU further uses the second position x ′ (t) and the second velocity vector v ′ (t) for the second virtual vehicle VEH 2 And calculate a second extrapolated position x '(t + L). The calculation of the extrapolation position is carried out, for example, based on the Runge-Kutta method, preferably the Euler method, preferably based on only one integration step over the estimated latency L. If the extrapolation position is too different from the actual position at time t + L, basically any higher precision integration method, eg higher order integration method, taking into account the relatively high computational complexity. Alternatively, multiple integrations may be used over partial intervals of latency L.

  The first arithmetic unit CPU substitutes the first extrapolation position x (t + L) and the second extrapolation position instead of the actual current first position x (t) and the second position x ′ (t). The position x ′ (t + L) is sent to the second operation unit GPU. Similarly, instead of another movable virtual object that may be present, ie, at least, instead of the position of the virtual object identified as having relevance with respect to the scenario modeled in the surrounding model In addition, the extrapolated position of the virtual object at all times is always sent. Thus, the second arithmetic unit GPU assumes, at the time of calculation of the image data, the estimated future state of the surrounding model MOD after the time period Δt has elapsed. When the image data calculated in this manner is finally supplied to the image processing system UUT, the simulation on the first arithmetic unit CPU substantially overcomes this time advantage of the image data. Thereby, the control data from the image processing system UUT better matches the current state M (t) of the surrounding model MOD, whereby the accuracy of the simulation results is better compared to similar testers known from the prior art Be improved.

Claims (15)

A testing machine for an image processing system (UUT),
A first computing unit (CPU) is disposed in the testing machine,
The first arithmetic unit (CPU) is programmed by simulation software for an ambient model (MOD),
The simulation software calculates a first position x (t) and a first velocity vector v (t), and the first position x (t) and the first velocity vector v (t) are It is configured to correspond to the first virtual object (VEH1) in the surrounding model (MOD),
A second computing unit (GPU) is disposed in the testing machine,
The second arithmetic unit (GPU) periodically reads the position of the first virtual object (VEH1) in the surrounding model, and based on at least the read position, the surrounding model (MOD) Configured to calculate first image data representing a two-dimensional first projection of
An adapter module (AD) is disposed in the testing machine,
The adapter module (AD) reads the first image data, processes the first image data by emulation of a first imaging sensor of the image processing system (UUT), and processes the first image data. Configured to provide image data to the image processing system (UUT);
The first arithmetic unit (CPU) reads control data calculated for the actuator by the image processing system (UUT) based on the processed first image data, and takes into account the control data. Are configured to associate a new first velocity vector with the first virtual object (VEH1),
In the test machine
The tester passes from the start of the calculation of the first image data by the second arithmetic unit (GPU) to the completion of the processing of the first image data by the adapter module (AD). Configured to measure the length Δt of the time interval,
The first arithmetic unit (CPU) is configured to read the length Δt of the time interval, and estimate the latency L of the first image data based on the length Δt of the time interval,
The first operation unit (CPU) takes into consideration the first position x (t) and the latency L estimated as the first velocity vector v (t). It is arranged to determine a first extrapolated position x (t + L) of the object, so that the first extrapolated position x (t + L) is of the first virtual object (VEH1) at time t + L. It is an estimate of the first position,
The second arithmetic unit (GPU) reads the first extrapolated position x (t + L) and calculates the first image data based on at least the first extrapolated position x (t + L). Is configured to
Testing machine characterized by The tester is configured to periodically calculate the first position x (t) and the first velocity vector v (t) in hard real time.
The testing machine according to claim 1. The first virtual object (VEH1) is a virtual vehicle.
The image processing system (UUT) is an automatic control or support system for vehicles.
The testing machine according to claim 1 or 2. Said first projection mimics the field of view of said first imaging sensor,
The tester according to any one of claims 1 to 3. The tester is configured to provide the first image data with a time stamp in which a first system time of the tester at the start of the calculation of the first image data is stored. ,
The adapter module (AD) reads the first system time stored in the timestamp and after completing the processing of the first image data, the time interval compared to the current system time To determine the length .DELTA.t and store the length .DELTA.t in a memory address,
The tester according to any one of claims 1 to 4. The tester creates a digital ID for the first image data, transmits the digital ID to the adapter module, and transmits the digital ID before the calculation of the first image data. Configured to send a first system time of the tester at the adapter module (AD),
The tester is configured to assign the digital ID to the first image data,
The adapter module (AD) is associating the first image data to said first system time based on said digital ID, after completing the processing of the first image data, the current of the tester The present system time is compared with the first system time to determine a length Δt of the time interval, and the length Δt is stored in a memory address.
The tester according to any one of claims 1 to 4. A first data connection (BS) having real time capability is formed between the first operation unit (CPU) and the second operation unit (GPU),
A second data connection (HDMI) having real time capability is formed between the second operation unit (GPU) and the adapter module (AD),
A real-time third data connection (ETH) is formed between the adapter module ( AD) and the first arithmetic unit.
The tester according to any one of claims 1 to 6. The first data connection is provided by a bus (BS) of the tester,
The testing machine according to claim 7. The second operation unit (GPU)
In parallel with the calculation of the first image data, or after the calculation of the first image data, at least two-dimensional of the surrounding model for a second imaging sensor of the image processing system (UUT) Calculating a second image data representing a second projection of the image, and creating a data packet including at least the first image data and the second image data,
The adapter module (AD) reads the data packet and processes the second image data by emulation of the second imaging sensor of the image processing system (UUT), the processed second image data Are configured to supply the image processing system (UUT),
A test machine according to any one of the preceding claims. The tester is configured to periodically determine the length Δt of the time interval,
The first arithmetic unit (CPU) is configured to periodically read the length Δt of the time interval,
The tester according to any one of claims 1 to 9. The first arithmetic unit (CPU) is configured to dynamically adapt the estimated latency L to the length of the time interval Δt by periodically setting L = Δt.
The testing machine according to claim 10. Said first arithmetic unit (CPU), by from a value of a plurality of Δt which is previously measured to calculate the value of the latency L, it is configured to dynamically adapt the pre Symbol estimated latency L ing,
The testing machine according to claim 10. The second arithmetic unit (GPU) is configured to selectively calculate the first image data for the different imaging sensors at least two respective,
The tester according to any one of claims 1 to 12. The simulation software calculates a second position x ′ (t) and a second velocity vector v ′ (t), and the second position x ′ (t) and the second velocity vector v ′ (t) ) Is configured to correspond to the second virtual object (VEH2) in the surrounding model,
The first calculation unit considers the second position x ′ (t), the second velocity vector v ′ (t), and the latency L estimated to be the second virtual unit. Configured to determine a second extrapolated position x ′ (t + L) of the object (VEH2),
The second arithmetic unit (GPU) reads the second extrapolation position x ′ (t + L), and at least the first extrapolation position x (t + L) and the second extrapolation position x ′ (t + L) Configured to calculate the first image data based on
The tester according to any one of claims 1 to 13. Using a test machine according to any one of the 請 Motomeko 1 to 14, a method for testing an image processing system (UUT),
The first arithmetic unit (CPU) of the test machine is programmed by simulation software for the ambient model and the simulation software cyclically generates the first position x (t) and the first velocity vector v (t) in hard real time. Calculated and associated with the first virtual object (VEH1) in the surrounding model,
The position of the first virtual object (VEH1) in the surrounding model is periodically read by the second operation unit (GPU) of the test machine, and is read by the second operation unit (GPU). Calculating first image data representing a two-dimensional first projection of the surrounding model based on the first position x (t);
Reading said first image data by an adapter module (AD) and processing by emulation of a first imaging sensor of said image processing system (UUT),
Supplying the image data processed by the adapter module (AD) to the image processing system (UUT);
The control data calculated for the actuator by the image processing system (UUT) is read by the first arithmetic unit (CPU) based on the processed first image data, and the control data is taken into account. To associate a new first velocity vector with the first virtual object (VEH1),
In the method
A length of time interval elapsed from the start of the calculation of the first image data by the second arithmetic unit (GPU) to the completion of the processing of the first image data by the adapter module (AD) Measure Δt,
The latency L of the first image data is estimated based on the length Δt of the time interval,
Taking into account the first position x (t) and the first velocity vector v (t) and the estimated latency L, a first extrapolated position x (t + L) is determined. Of the first virtual object (VEH 1 ) at time t + L, the extrapolated position x (t + L) of the first virtual object at time t + L,
Calculating the first image data by the second operation unit (GPU) based on the first extrapolated position x (t + L);
A method characterized by

Images