Environment perception system for a vehicle

Abstract

The invention relates to an environment perception system (1) for a vehicle comprising a detection module (2) which scans and models the environment in order to produce a first stream (3) of point clouds, the perception system (1) comprising an adaptation module (4) for adapting to the meteorological conditions of the environment which is able to modify the point clouds of the first stream (3), and a driving module (6) comprising a first member (8) which is able to operate the vehicle in a conventional driving mode (MC) and a second member (10) which is able to operate the vehicle in a degraded driving mode (MD), characterized in that the adaptation module (4) comprises a first analysis device (12) for analyzing at least one raw cloud (24) of the first stream (3) and a filtration device (14) which is able to switch between an active mode (ON) in which the filtration device (14) filters the point clouds of the first stream (3) in order to produce a filtered stream (16) of filtered clouds, and an inactive mode (OFF) in which the filtration device (14) leaves the first stream (3) unchanged and delivers it to the first member (8) of the driving module (6), the adaptation module (4) comprising at least one second analysis device (18) configured to analyze at least one filtered cloud (34) of the filtered stream (16) and deliver the filtered cloud (34) of the filtered stream (16) to the first member (8) of the driving module (6) or to the second member (10) of the driving module (6) according to information relating to the filtered cloud (34).

Description

Description Title of the invention: Environmental perception system for a vehicle

[0001] The present invention relates to the field of vehicles, and more particularly to systems for perceiving an environment around vehicles.

[0002] Modern vehicles are often equipped with systems that perceive the environment surrounding them. These perception systems can range from a simple reversing radar on a conventional vehicle to more advanced systems for guiding autonomous vehicles.

[0003] Perception systems include sensing modules, such as global positioning systems (GPS) or laser-based remote sensing (LiDAR). In adverse weather conditions, such as rain, snow, fog, or dust, the capabilities of these sensing modules can be impaired. These conditions can interfere with environmental perception, leading to outliers in the data produced by the sensing modules.

[0004] The presence of outliers can lead to the detection of phantom objects or, conversely, the failure to detect real objects. This is problematic because incorrect detection could lead to dangerous driving, compromising safety due to an inaccurate perception of the environment.

[0005] To eliminate outliers, some prior art technologies use special detection modules, such as modules that dynamically modify their parameters or modules with separate emitters and receivers. Furthermore, some of these technologies use multiple detection modules to compare data and remove outliers. However, these solutions require modifications to the detection modules or the use of multiple detection modules, which can be complex and expensive.

[0006] Other technologies operate at the software level, removing outliers after data collection by the detection module, but they do not guarantee complete safety through multiple data analyses. Furthermore, some of these technologies often focus on outliers related to road conditions rather than those related to atmospheric weather conditions.

[0007] The objective of the invention described in this document is therefore to overcome the disadvantages of the prior art by eliminating, at the software level, the aberrations due to adverse weather conditions, particularly atmospheric ones.

[0008] The invention thus overcomes the limitations of the prior art by using a commercially available standard detection module and by including various analysis mechanisms ensuring increased safety for the vehicle.

[0009] The main object of the present invention is a vehicle environmental perception system comprising a detection module that scans and models said environment in order to produce a first stream of point clouds, the perception system comprising a weather-condition adaptation module for said environment capable of modifying the point clouds of said first stream, and a driving module comprising a first component capable of operating the vehicle in a conventional driving mode and a second component capable of operating the vehicle in a degraded driving mode, characterized in that the adaptation module comprises a first device for analyzing at least one raw cloud from the first stream and a filtering device capable of switching between an active mode in which the filtering device filters the point clouds of the first stream in order to produce a filtered stream of filtered clouds,and an inactive mode in which the filtration device leaves the first stream unchanged and delivers it to the first component of the control module, the adaptation module comprising at least one second analysis device configured to analyze at least one filtered cloud of the filtered stream and deliver said filtered cloud of the filtered stream to the first component of the control module or to the second component of the control module according to information relating to said filtered cloud.

[0010] The invention presented relates to all types of vehicles using a detection module as a primary or secondary source of information about their environment. For example, the detection module can use LiDAR technology to scan and model the environment as a point cloud. It should be noted that each scan creates a point cloud modeling the environment at a given time t, and that scans are performed multiple times to obtain a stream of point clouds, referred to herein as the first stream.

[0011] The first stream is then directed to the adaptation module, which modifies it in case of adverse weather conditions that cause outliers in the point clouds. To do this, the adaptation module uses the filtration device, which, when active, can apply different methods to filter each point cloud of the first stream, i.e., remove the outliers. The filtration device then produces filtered clouds that form the filtered stream. In inactive mode, the filtration device allows the point clouds to pass through without modification, directing them directly to the first component of the driving module for use in conventional driving mode.

[0012] The first analysis method examines at least one point cloud from the first stream; this cloud is called the raw cloud because it has not been filtered. This analysis aims to The first analytical device determines whether the raw cloud contains outliers, which allows for the inference of weather conditions in the vehicle's environment. Indeed, when the raw cloud contains outliers, the first analytical device assesses whether these outliers are due to adverse weather conditions and, if so, infers the presence of such conditions.

[0013] In active mode, the filtration system sends the filtered stream to the second analysis unit, which checks the quality of the filtered data points. It does this by analyzing one data point at a time, checking two things: first, whether the filtered data point contains enough data points, and second, whether any outliers remain. If the filtered data point lacks sufficient data points or still contains outliers, it is sent to the second unit for use in degraded driving mode. Conversely, if the filtered data point is of good quality, it is sent to the first unit for normal driving mode.

[0014] The standard driving mode corresponds to normal vehicle operation with all its features. The degraded driving mode, on the other hand, is a simplified version designed to ensure safe driving in exchange for reduced functionality. For example, this mode can reduce the vehicle's speed and focus on point clouds corresponding to the immediate surroundings, as these are more likely to reliably detect obstacles and brake in an emergency. If there are too many outliers or an insufficient number of points in the clouds reaching the second organ, the degraded driving mode can also choose to bring the vehicle to a complete stop to avoid accidents.

[0015] According to an optional feature of the invention, the filtration device is placed under the dependence of the first analysis device.

[0016] It is thus understood that the filtration device switches to active mode or remains in inactive mode depending on the analysis of the raw cloud made by the first analysis device.

[0017] During analysis, when the first analysis device concludes that the raw cloud contains outliers corresponding to adverse weather conditions, the perception system concludes that adverse weather conditions are affecting the environment and, consequently, the point clouds in the first stream. The first analysis device sends a signal, called the "true" signal, to activate or maintain the filtering device in active mode.

[0018] Conversely, if adverse weather conditions are undetectable in the raw cloud, it sends a signal, called a "false" signal, to switch or keep the filtration device in inactive mode.

[0019] This feature allows the filtration system to operate only when adverse weather conditions are present in the environment. This allows for considerable resource savings by filtering only the point clouds from the first stream containing outliers.

[0020] According to an optional feature of the invention, the perception system is configured so that the first analysis device and the filtration device simultaneously receive the first flow.

[0021] It is thus understood that the first analysis device and the filtering device are placed in parallel to receive the first stream. The filtering device allows or filters the point clouds of the first stream, while the first analysis device analyzes the raw point cloud of this first stream.

[0022] Receiving the first stream simultaneously prevents the first analysis device from receiving it before the filtering device, which could lead to wasted resources. This is because when the first analysis device is positioned before the filtering device, each point cloud, whether clean or affected by adverse weather conditions, must first be analyzed and then sent to the filtering device. Consequently, for each point cloud in the first stream that is not affected by adverse weather conditions, the first analysis device wastes resources searching for those conditions.

[0023] According to an optional feature of the invention, the first analysis device comprises a first counting means and a first analysis means, the first counting means counting the number of point clouds of the first stream over a given period, said first analysis means performing an analysis of the raw cloud of the first stream when the number of point clouds counted by the first counting means reaches a cloud threshold.

[0024] The first counting element is the component of the first analysis device that receives the first stream and thus counts each of the point clouds in the first stream. It is important to understand that, since the first analysis device and the filtering device are placed in parallel, the point clouds of the first stream arrive at both components simultaneously.

[0025] The first counting method therefore counts the number of point clouds in the first stream until it exceeds a defined threshold. Once this threshold is reached, the point cloud exceeding this threshold is passed to the first analysis method. This cloud is called the raw cloud and is then analyzed by the first analysis method.

[0026] Using this counting method, a raw cloud of the first stream is periodically analyzed. For example, if the cloud threshold is 600, the first analysis method analyzes a raw cloud of the first stream every 600 point clouds arriving at the first counting method and thus at the filtering device.

[0027] This periodic analysis saves resources, because the first means of analysis, and therefore the first analytical device, only analyzes a raw cloud periodically, thus avoiding the analysis of each point cloud in the first stream. This method is considered safe since adverse weather conditions take some time to develop.

[0028] According to an optional feature of the invention, the first analysis means is configured to control the filtration device by comparing an intensity of points of said raw cloud with an intensity threshold.

[0029] Here, we utilize the detection module's ability to determine the intensity of points in point clouds. It's important to note that outliers caused by weather conditions have lower intensity than so-called normal points. Therefore, if the intensity of a point in the raw cloud is below the intensity threshold, it is considered an outlier.

[0030] It should be noted that the first analysis method must then determine whether this outlier is due to adverse weather conditions. To do this, it can, in particular, use the distribution of outliers: a disordered distribution could thus be a sign of adverse weather conditions. The first analysis method then classifies the raw cloud as either affected or unaffected by adverse weather conditions and then sends the signal "true" or "false," respectively, to the filtering device.

[0031] According to an optional feature of the invention, the first counting means is reset to zero each time the first analysis means performs an analysis.

[0032] This feature allows the first counting method to reset itself and start counting again until it reaches the cloud threshold once more. Thus, the first analysis method periodically performs an analysis on a raw cloud of the first stream.

[0033] According to an optional feature of the invention, the filtering device uses a statistical outlier removal algorithm or dynamic statistical outlier removal to filter the point clouds of the first stream.

[0034] These methods analyze the point clouds of the first stream point by point, searching for points that are spatially isolated from other points. Then, they remove the identified outliers, forming a filtered cloud that is normally free of outliers.

[0035] According to an optional feature of the invention, the detection module is a laser remote sensing device.

[0036] Laser-based remote sensing (LiDAR) devices scan their environment by emitting laser pulses that are reflected by objects and captured by the LiDAR device, thus forming data packets. To model the environment, these data packets are then read, decrypted and converted into a point cloud using calibration files.

[0037] Some LiDARs can be 360-degree LiDARs, performing a complete rotation of their lasers around their axis. It should be noted that any type of these LiDAR devices can be used, for example, LiDAR devices operating at 10 Hz. Operating at 10 Hz means that the LiDAR devices perform a complete scan 10 times per second and therefore generate a new point cloud every 0.1 seconds. Thus, if the cloud threshold is set to 600, a raw point cloud is analyzed every 60 seconds.

[0038] According to an optional feature of the invention, the second analysis device comprises a second counting means and a second analysis means, the second counting means counting the number of points within the filtered cloud of the filtered stream, said second analysis means performing an analysis of the filtered cloud when the number of points within said filtered cloud is greater than a point threshold.

[0039] The second analysis device adds a safeguard to ensure that the filtered clouds from the filtered stream are usable by the first component in normal driving mode. Indeed, in severe adverse weather conditions, the filtering device may eliminate not only outliers but also normal ones, and it may fail to remove all outliers. Therefore, it is possible to obtain filtered clouds from the filtered stream that lack information or still contain outliers.

[0040] The second counting method counts the number of points in each filtered data point cloud of the filtered data stream. Each filtered data point cloud has its points counted and compared to the specified point threshold. When the number of points is below the threshold, it means there are insufficient points in the filtered data point cloud. Conversely, when the number of points is above the threshold, it indicates that there are enough points in the filtered data point cloud for it to be usable in normal driving mode. This filtered data point cloud is then sent to the second analysis device to check for any remaining outliers.

[0041] According to an optional feature of the invention, the second analysis means is configured to deliver the filtered cloud to the first organ of the driving module or to the second organ of the driving module depending on the result of a comparison between an intensity of the points of said filtered cloud and an intensity level.

[0042] The second analysis method checks for any remaining outliers in the filtered cloud to determine where it should be directed. This method works similarly to the first analysis method, where points are considered outliers if they are below the intensity threshold, and normal points if they are above it.

[0043] Furthermore, the second analysis method also uses point distribution to determine if outliers are due to weather conditions. When so, it sends these outliers to the second system for degraded driving mode.

[0044] According to an optional feature of the invention, when the number of points in the filtered cloud is less than the point threshold, the second counting means delivers the filtered cloud to the second component of the driving module. In this way, filtered clouds with an insufficient number of points are handled by the second component for the degraded driving mode.

[0045] The invention also relates to a method for controlling a vehicle's driving mode, comprising a perception system as previously described. More specifically, according to the invention, a method for controlling a driving mode comprises:

[0046] - a first step in which the detection module scans and models the environment in order to produce the first stream of point clouds,

[0047] - a second step in which the first analysis device of the adaptation module analyzes the raw cloud of the first flow and then sends a signal causing the filtering device to switch between active and inactive modes:

[0048] - an optional third step in which, when the filtration device is in active mode, the filtered cloud of the filtered flow is delivered according to information relating to the filtered cloud to the first organ or the second organ of the driving module.

[0049] Depending on an optional feature, the second step in the vehicle driving mode control process includes:

[0050] a first sub-step in which the first counting method counts the number of point clouds in the first stream over a given period,

[0051] a second sub-step in which, when the number of point clouds counted by the first counting means reaches the cloud threshold, the first analysis means performs an analysis of the raw cloud of the first flow by comparing the intensity of points of said raw cloud with the intensity threshold and then controls the switching of the filtering device between active mode and inactive mode according to the result of this analysis,

[0052] a third sub-step in which the first counting method is reset to zero.

[0053] Depending on an optional feature, the third optional step in the vehicle driving mode control process includes:

[0054] a first sub-step in which the second counting method counts the number of points within the filtered cloud of the filtered stream,

[0055] a second sub-step in which:

[0056] - when the number of filtered points in the cloud is less than the point threshold, the second counting means delivers said filtered cloud to the second component of the control module,

[0057] - when the number of points within said filtered cloud is greater than the point threshold, the second analysis means performs an analysis in which the second analysis device delivers the filtered cloud to the first organ of the control module or to the second organ of the control module depending on the result of a comparison between the intensity of the points of said filtered cloud and the intensity level.

[0058] Other features, details and advantages of the invention will become clearer upon reading the following description on the one hand, and the illustrative and non-limiting examples of embodiments given with reference to the accompanying drawings on the other hand, in which:

[0059] [Fig.1] is a schematic representation of an environmental perception system for a vehicle.

[0060] [Fig.2] is a schematic representation of a first analysis device and a filtering device of an adaptation module of the perception system.

[0061] [Fig.3] is a schematic representation of a second analysis device for the adaptation module.

[0062] The features and variants of the invention can be combined in various ways, provided they are not incompatible or mutually exclusive. In particular, variants of the invention may be conceived comprising only a selection of the features described below, isolated from the other described features, if this selection of features is sufficient to confer a technical advantage and / or to differentiate the invention from the prior art.

[0063] In the figures, elements common to several figures retain the same reference.

[0064] Fig. 1 is a schematic representation of a vehicle's environmental perception system.

[0065] This perception system 1 can notably be used in autonomous vehicles, whether they are mobile robots, cars or autonomous land vehicles.

[0066] The perception system 1 includes, in particular, a detection module 2 that scans and models the environment surrounding the perception system 1. To do this, the detection module 2 can use laser remote sensing (LiDAR) technology. The detection module 2 then scans its environment using an array of lasers and detectors, sending out laser pulses and by measuring the delay, also called "time of flight", between the emission of laser pulses and the detection of these pulses reflected by an object.

[0067] The measurements of the emitted laser pulses that bounced off objects and were then captured again by detection module 2 form a data block. This data block is then read and, using calibration files, decoded and converted into a 3D point cloud.

[0068] The detection module 2 operates here at 10 hertz, producing a point cloud every 0.1 seconds, or 10 clouds per second. It is therefore clear that the detection module 2 produces a plurality of point clouds, forming a first stream 3 of point clouds. It should be noted that detection modules operating at other hertz values ​​can be used within the invention.

[0069] The detection module 2 captures information here via 360° scans, that is, by rotating the lasers 360 degrees around an axis, which makes it possible to obtain point clouds representing the environment in 360° around the perception system 1.

[0070] In the event of rain, snow, or fog—that is, adverse weather conditions—the point clouds of the first stream 3 may contain outliers resulting from the collision of the lasers of the detection module 2 with raindrops, snow, hail, or fog. If these outliers are not addressed and eliminated, they can lead to problems in the operation of the vehicle on which the detection module 2 is positioned, such as detecting objects where none exist or failing to detect actual objects.

[0071] Perception system 1 has the unique ability to adapt to various weather conditions. To this end, it includes an adaptation module 4 that resolves problems arising from adverse weather conditions by identifying and filtering outliers from the point clouds of the first stream (3).

[0072] The perception system 1 also includes a driving module 6 comprising a first component 8 which can operate the vehicle in a conventional driving mode MC when it receives a stream of point clouds and a second component 10 which can operate the vehicle in a degraded driving mode MD when it receives a stream of point clouds.

[0073] The MD Degraded Driving Mode is a simplified version of the MC Classic Driving Mode. It includes simpler versions of the main driving controllers, designed to take control and / or assist vehicle functions in adverse weather conditions. This MD Degraded Driving Mode allows for safe navigation in exchange for reduced functionality. and the vehicle's capabilities. For example, the degraded driving mode can halve the vehicle's speed and focus on readings from the six closest detection modules, as these are more likely to be reliable. This allows for obstacle detection and emergency braking, due to the vehicle's low speed in this MD driving mode.

[0074] It should be noted that the adaptation module 4 is positioned between the detection module 2 and the driving module 6, and that it allows a stream of point clouds to be sent either to the first organ 8 or to the second organ 10 in order to choose between the classic driving mode and the degraded driving mode.

[0075] The detection module 2 includes, in particular, a first analysis device 12 that analyzes at least one point cloud from the first stream 3, called the raw point cloud, in order to detect outliers in the first stream 3 of point clouds. The detection module 2 also includes a filtering device 14 for outliers in the point clouds of the first stream 3.

[0076] The first analysis device 12 periodically analyzes the raw cloud of the first stream 3 to identify any outliers induced by weather conditions. The filtering device 14 can be activated in an active mode (ON) or deactivated in an inactive mode (OFF) as needed. After the periodic analysis, the first analysis device 12 sends a signal 14 that switches the filtering device 14 from the active mode (ON) to the inactive mode (OFF) and vice versa.

[0077] When in active ON mode, the filtering device 14 operates by processing the point clouds of the first stream 3 generated by the detection means 2 in order to eliminate outliers that do not represent real objects or obstacles and therefore correspond to weather conditions. It then produces a filtered stream 16 of filtered point clouds.

[0078] When in inactive OFF mode, the filtration device 14 allows the first stream 3 to pass through unfiltered, freeing up resources for other functions. In this case, the first stream 3 exits the filtration device 14 unchanged and is directly transmitted to the drive module 6. Since the first stream 3 is unchanged and has therefore been considered by the system as a cloud stream without outliers, it is given to the first component 8 of the drive module 6 to utilize its point clouds in the vehicle's conventional driving mode (MC).

[0079] The filtration device 14 and the first analysis device 12 receive the first stream 3 simultaneously in order to increase the efficiency of the perception system 1. Indeed, since the perception system 1 was specifically designed for autonomous vehicles and for deployment in an embedded computer, the processing must to be carried out in real time, where every millisecond counts and where one must seek to conserve computing resources.

[0080] Indeed, it should be noted that if the first analysis device 12 receives the first stream 3 before the filtering device 14, the first stream 3 must first be analyzed in its entirety and then sent to the filtering device 14. Therefore, for each cloud of the first stream 3 that is not affected by outliers, the perception system 1, and more specifically the first analysis device 12, loses resources and time searching for outliers.

[0081] The filtered stream 16 of filtered point clouds from the filtering device 14 is then analyzed by a second analysis device 18 of the adaptation module 4.

[0082] The second analysis device 18 provides greater robustness and security for the perception system 1. Indeed, after the filtering device 14, in rare cases, adverse weather conditions may be so severe that the filtering device 14 removes most of the points, leaving too few points in the filtered point clouds of the filtered stream 16 to guarantee reliable object detection. This can occur, for example, in dense fog. It is also possible that the filtering device 14 may not have been able to remove all the outliers from the point clouds of the first stream 3, resulting in a filtered stream 16 containing filtered point clouds that still include outliers. In this case, even after the filtering device 14, the filtered point clouds of the filtered stream 16 are unusable in conventional driving mode (MC).

[0083] The second analysis device 18 therefore analyzes the filtered clouds from the filtered stream 16 to ensure they are of sufficient quality for use by the driving module 6 in conventional mode. If there are still aberrant points in the filtered clouds from the filtered stream 16, or if there are not enough points in the filtered clouds from the filtered stream 16, these filtered clouds are then sent to the second component 10 of the driving module 6 to operate the vehicle in degraded driving mode MD. If the quality of the filtered clouds from the filtered stream 16 is sufficient, then these clouds are transmitted to the first component 8, which then operates the vehicle in conventional driving mode MC.

[0084] Fig. 2 is a schematic representation of the first analysis device 12 and the filtering device 14 of the adaptation module 4 of the perception system 1.

[0085] It should be noted that the only input to the first analysis device 12 is the first stream 3.

[0086] The first analysis device 12 includes a first counting means 20 and a first analysis means 22. Each time a point cloud from the first stream 3 reaches the first analysis device 12, the first counting means 20 increments an internal counter by 1, thus counting the number of point clouds in the first stream 3 reaching it. Once this counter reaches a desired number, called here the cloud threshold SN, which corresponds to a number of clouds from the first stream 3 counted, the first analysis means 22 is activated and analyzes a cloud from the first stream 3, called the raw cloud and referenced 24 on the [Fig.2].

[0087] The cloud threshold SN of the first counting medium 20 is configurable and can be adjusted according to the environment around the vehicle. A cloud threshold SN = 1 means that every cloud in the first stream 3 will be analyzed for adverse weather conditions, while a cloud threshold SN = 600 means that only one cloud out of 600 is analyzed. It follows that the lower the cloud threshold SN, the more frequently the first analysis medium 22 analyzes a raw cloud 24 from the first stream 3 for weather-induced outliers; however, the first analysis medium 22 becomes computationally and resource-intensive. The cloud threshold SN is therefore configured to achieve a compromise between protection against sudden weather changes and avoiding excessive resource consumption.Thus, according to the inventors, for a case of use of an outdoor vehicle for logistics, a point cloud threshold SN = 600, which corresponds, when using a 10 hertz detection module 2, to an analysis of a raw cloud 24 of the first flow 3 every 60 seconds, proved to be a reliable threshold, since weather conditions tend to develop slowly.

[0088] When the SN cloud threshold is reached, the first analysis means 22 performs an analysis on the raw cloud 24 of the first flux 3 that exceeded the SN cloud threshold counted by the first counting means 20. This is done by exploiting the ability of the detection module 2 to obtain the intensity of lasers that have bounced off an object and are captured by the detection module 2. It should be noted that lasers impacting various weather conditions, such as fog or raindrops, return with very low intensity, since these weather conditions scatter most of the light emitted by the detection module 2 instead of simply reflecting it. Consequently, outliers are actually points with a lower intensity than the other points.Thus, to detect outliers, the first analysis method 22 compares in particular the intensity of each point of the raw cloud 24 against an intensity threshold of the SL points. However, other measured objects may also have low intensity, such as extremely dark surfaces or tinted glass.

[0089] Therefore, the first analysis method 22 also performs a study of the distribution of low-intensity points to determine whether the outliers are due to real objects or meteorological conditions. The first analysis method 22 then performs a classification using this point intensity and distribution, which determines whether the raw cloud 24 is affected or not. due to adverse weather conditions. It should be noted that for this classification, the first analysis method 22 can use any classifier as long as it is correctly configured. The first analysis method 22 can also use any other factors to classify the raw cloud 24 to determine if there are any outliers due to adverse weather conditions.

[0090] Once the analysis by the first analysis method 22 is complete, and following the classification performed on the raw cloud 24, it sends the signal 14 to the filtering device 14. More precisely, it sends a "true" signal 25 when the raw cloud 24 is polluted by outliers due to weather conditions, and a "false" signal 26 when the raw cloud 24 is not polluted. It is thus understood that the "true" signal 25 activates the filtering device 14 by switching it to or keeping it in active ON mode, and the "false" signal 26 deactivates the filtering device 14 by switching it to or keeping it in inactive OFF mode.

[0091] When the analysis of the first analysis means 22 is completed, the first analysis means 22 also sends a reset signal 28 to the first counting means 20, which allows the first counting means 20 to reset itself and thus restart the counting of the point clouds of the first stream 3 until they again reach the cloud threshold SN and an analysis by the first analysis means 22 of a raw cloud 24 of the first stream 3 starts again.

[0092] In parallel with the first analysis device 12, the filtration device 14 therefore filters the first flow 3 depending on whether it is in active mode ON or in inactive mode OFF.

[0093] The filtering device 14 comprises two inputs: one input through which the first stream 3 arrives, and one input through which the "true" signal 25 or the "false" signal 26 arrives. More precisely, upon arrival of the "true" signals 25 or "false" 26, these signals are recorded in a signal memory of the filtering device 14. Thus, when the filtering device 14 is in active ON mode, the memory has recorded the "true" signal 25, whereas if it is in inactive OFF mode, the memory has recorded the "false" signal 26. The switch from active ON mode to inactive OFF mode occurs when the signal changes from "true" to "false" and vice versa, and then the signal is recorded by the memory.

[0094] As soon as a point cloud from the first stream 3 arrives at the filtering device 14 (ten per second when the detection module 2 is at 10 hertz), the filtering device 14 accesses its signal memory to determine whether a "true" signal 25 or a "false" signal has been recorded, and thus whether the filtering device 14 is in active ON mode or inactive OFF mode. When the filtering device 14 is in inactive OFF mode, and therefore the memory has recorded the "false" signal 26, it allows the first stream 3 to flow without modification and delivers it to the control module. 6. It is therefore understood that if there are no adverse weather conditions, the filtration device 14 acts only as a passage for the first flow 3. It therefore does not perform any operation and leaves the point clouds of the first flow 3 unchanged, which avoids wasting time and resources.

[0095] When the filtering device 14 is in active mode ON, i.e. the memory has recorded the "true" signal 25, then the filtering device 14 filters each of the point clouds of the first stream 3 until the filtering device 14 returns to inactive mode, i.e. the weather conditions have improved.

[0096] To remove outliers caused by weather conditions, the filtration device 14 can use several methods. Techniques such as statistical outlier removal (SOR) algorithms or dynamic statistical outlier removal (DSOR) can be used.

[0097] These methods analyze the points in the point clouds of the first stream 3 point by point, searching for statistically outliers, that is, points that are spatially isolated from the others. They then remove these identified outliers, leaving the point clouds of the first stream 3 free of outliers. Consequently, the filtering device 14, when the active mode ON is activated, thus forms the filtered stream 16 of filtered clouds without outliers. It should be noted, however, that in rare cases, some outliers may still be present within the filtered clouds; therefore, the filtered stream 16 is then delivered by the filtering device 14 to the second analysis device 18.

[0098] The filtering by the filtering device 14 is performed very quickly, in well under 100 milliseconds. Thus, the filtered clouds of the filtered stream 16 are generated at the same rate as the point clouds of the first stream 3, i.e. 10 hertz if a detection module 2 operating at 10 hertz is used.

[0099] It is worth noting that switching the filtration device 14 between active ON and inactive OFF modes is particularly advantageous. This is because the filtration methods are computationally intensive, as they analyze the point clouds of the first three streams point by point. Therefore, without this function of enabling or disabling the filtration device 14, it would operate unnecessarily, for example, on sunny days, resulting in a considerable waste of resources.

[0100] Fig. 3 is a schematic representation of the second analysis device 18 of the adaptation module 4.

[0101] This second analysis device 18 is designed to verify that the filtration device 14 has functioned correctly and to determine whether the filtered clouds of the filtered flow 16 are usable for driving the vehicle in conventional driving mode. MC, or if the vehicle must be driven in degraded driving mode MD to avoid potential hazards. Indeed, the filtered clouds of the first stream 3 could be so polluted, for example in dense fog, that the filtering device 14 would be unable to remove all the outliers, rendering the filtered clouds of the second filtered stream 16 unusable for the vehicle's normal driving mode MC. Similarly, if the filtered clouds of the first stream 3 are significantly polluted, the filtering device 14 might correctly remove all outliers, but leave the filtered clouds of the second filtered stream 16 with too few points to extract relevant information, which could lead to considerable hazards in normal driving mode MC.

[0102] It should be noted that the second analysis device 18 has a relatively short analysis time. Therefore, analyzing the filtered clouds of the filtered stream 16 does not significantly delay the information flow.

[0103] The sole input to the second analysis device 18 is the filtered stream 16. Therefore, when no adverse weather conditions are detected by the first analysis device 12, and the filtering device 14 does not generate a filtered stream 16, the second analysis device 18 is inactive. However, if adverse weather conditions are detected by the first analysis device 12, and the filtering device 14 filters the point clouds of the first stream 3, then the second analysis device 18 analyzes the filtered stream 16.

[0104] In order to verify the filtration performed by the filtration device 14, i.e. the quality of the filtered stream 16, the second analysis device 18 includes a second counting means 30 and a second analysis means 32. The second counting means 30 operates differently from the first counting means 20. Indeed, unlike the first counting means 20 which counts the filtered clouds, the second counting means counts the points of each filtered cloud in the filtered stream 16.

[0105] Thus, when a filtered cloud from the filtered stream 16 arrives at the second counting means 30 of the second analysis device 18, the latter scans the filtered cloud and studies its point count. If this number is less than a point threshold SP, defined to ensure that there is sufficient data for the perception system 1 to detect all surrounding obstacles, the second counting means 30 diverts the filtered cloud with an insufficient number of points to the second component 10 of the driving module 6, so that it can be used in degraded driving mode. This filtered cloud is represented by reference 34 in [Fig. 3].

[0106] When the number of points counted by the second counting method 30 exceeds the defined SP point threshold, it directs the filtered cloud 34 with a number sufficient points towards the second analysis means 32 of the second analysis device 32.

[0107] The second analysis means 32 operates in a relatively similar manner to the first analysis means 22 of the first analysis device 12. The second analysis means 32 examines the filtered cloud 34 for outliers, using the ability of the detection module 2 to determine the intensity of lasers that have reflected off an object. To detect these outliers, the second analysis means 32 compares the intensity of each point in this filtered cloud 34 to a point intensity level NI. The second analysis means 32 also studies the distribution of low-intensity points in this filtered cloud 34.

[0108] Next, the second analysis means 32 performs a classification using this point intensity and distribution, which determines whether the filtered cloud 34 is still affected by adverse weather conditions. It should be noted that for analysis and classification, the second analysis means 32 can use the same types of means as those used by the first analysis means 22 of the first analysis device 12.

[0109] When, at the end of the classification, the second analysis means 32 classifies the filtered cloud 34 as a filtered cloud 34 without outliers, then the second analysis device 18 directs this filtered cloud 34 without outliers to the first organ 8 for use in the vehicle's classic MC driving mode.

[0110] When outliers are still detected in the filtered cloud 34, it is then considered a filtered cloud 34 with outliers. This filtered cloud 34 with outliers is then directed to the second component 10 for use in the vehicle's MD degraded driving mode.

[0111] The invention also relates to a method for controlling the driving modes of said vehicle, which uses the perception system 1.

[0112] The process includes a first step in which the detection module 2 scans and models the environment to produce the first stream 3 of point clouds. When the detection module 2 uses LiDAR technology, it employs multiple lasers and sensors to calculate the time of flight of the lasers and produce the points of the point clouds of the first stream 3.

[0113] The process also includes a second step in which the first analysis device 12 examines the raw cloud 24 of the first stream 3, and then sends the signal 14, enabling it to switch to or remain in inactive OFF mode or active ON mode. It should be understood that the signal 14 can be either the "true" signal 25, which causes the filtering device 14 to switch to or remain in active ON mode, or the "false" signal 26, which causes the device to switch to or remain in inactive OFF mode.

[0114] This second step may include a first substep in which the first counting means 20 counts the number of point clouds from the first stream 3 over a period defined by the number of point clouds arriving at the first counting means 20 and by the cloud threshold SN.

[0115] The second step may also include a second sub-step and a third sub-step. In the second sub-step, when the number of point clouds counted by the first counting means 20 reaches the cloud threshold SN, the first analysis means 22 compares the intensity of the points in the raw cloud 24 with the intensity threshold SI, and then controls the switching of the filtering device 14 between the active mode ON and the inactive mode OFF according to the result of this analysis, by means of the emission of the "true" signals 25 or "false" 26.

[0116] In the third sub-step of the second step, the first counting means 20 is reset to zero when the analysis by the first analysis means 22 is completed.

[0117] The process also includes an optional third step following the second step. In this optional third step, when the filtration device 14 is in active mode ON, the filtered cloud 34 of the filtered stream 16 is delivered according to information relating to the filtered cloud 34, either to the first organ 8 which operates the vehicle in conventional driving mode MC, or to the second organ 10 of the driving module 6 which operates the vehicle in degraded driving mode MD.

[0118] To perform this optional third step, the third step may include a first substep and a second substep. In this first substep, the second counting method 32 counts the number of points within the filtered cloud 34 of the filtered stream 16.

[0119] In the second sub-step of the optional third step, when the number of points in the filtered cloud 34 counted by the second counting means 30 is less than the SP point threshold, the second counting means 30 delivers it to the second organ 10 of the driving module 6.

[0120] When the number of points within the filtered cloud 34 counted by the counting means 30 exceeds the SP point threshold, the second analysis means 32 performs an analysis of said cloud and delivers it to the first component 8 of the control module 6 or to the second component 10 of the control module 6, based in particular on the result of the comparison between the intensity of the points of said filtered cloud 34 and the intensity level NI. Indeed, if the filtered cloud 34 is classified by the second analysis means 32, during this second sub-step of the optional third step, as a filtered cloud 34 without outliers, it is delivered to the first component 8, while if it is classified as a filtered cloud 34 with outliers, it is delivered to the second component 10.

[0121] This process thus makes it possible to guarantee efficient use of resources and safe driving of the vehicle by adapting data processing according to environmental conditions.

[0122] As just described, the present invention effectively achieves its objective by proposing a perception system comprising a detection module that scans and models the environment of a vehicle, an adaptation module that analyzes weather conditions and filters information from adverse conditions, and a driving module capable of using either a degraded driving mode or a classic driving mode depending on the information provided by the detection module.

[0123] The present invention is not limited to the means and configurations described and illustrated herein, and also extends to any equivalent means and configuration as well as any technically operative combination of such means.

Claims

Demands

1. A vehicle environmental perception system (1) comprising a detection module (2) scanning and modeling said environment to produce a first stream (3) of point clouds, the perception system (1) comprising an adaptation module (4) to the weather conditions of said environment capable of modifying the point clouds of said first stream (3), and a driving module (6) comprising a first component (8) capable of operating the vehicle in a conventional driving mode (CM) and a second component (10) capable of operating the vehicle in a degraded driving mode (MD), characterized in that the adaptation module (4) comprises a first analysis device (12) for at least one raw cloud (24) of the first stream (3) and a filtering device (14) capable of switching between an active mode (ON) in which the filtering device (14) filters the point clouds of the first stream (3) to produce a filtered stream (16) of filtered clouds,and an inactive (OFF) mode in which the filtering device (14) leaves the first stream (3) unchanged and delivers it to the first component (8) of the control module (6), the adaptation module (4) comprising at least one second analysis device (18) configured to analyze at least one filtered cloud (34) of the filtered stream (16) and deliver said filtered cloud (34) of the filtered stream (16) to the first component (8) of the control module (6) or to the second component (10) of the control module (6) according to information relating to said filtered cloud (34).

2. Perception system (1) according to claim 1, wherein the filtration device (14) is placed under the dependence of the first analysis device (12).

3. A perception system according to any one of claims 1 or 2, configured so that the first analysis device (12) and the filtration device (14) simultaneously receive the first stream (3).

4. A perception system (1) according to any one of claims 1 to 3, wherein the first analysis device (12) comprises a first counting means (20) and a first analysis means (22), the first counting means (20) counting the number of point clouds of the first stream (3) over a given period, said first analysis means (22) performing an analysis of the raw cloud (24) of the first stream (3) when the number of point clouds counted by the first counting means (20) reaches a threshold of SN clouds.

5. Perception system (1) according to claim 4, wherein the first analysis means (22) is configured to control the filtering device (14) by comparing an intensity of points of said raw cloud (24) with an intensity threshold (SI).

6. Perception system (1) according to any one of claims 4 or 5, wherein the first counting means (20) is reset to zero each time the first analysis means (22) performs an analysis.

7. Perception system (1) according to any one of claims 1 to 6 wherein the filtering device (14) uses a statistical outlier removal algorithm or dynamic statistical outlier removal in order to filter the point clouds of the first stream (3).

8. Perception system (1) according to any one of claims 1 to 7, wherein the detection module (2) is a laser remote sensing device.

9. Perception system (1) according to any one of claims 1 to 8, wherein the second analysis device (18) comprises a second counting means (30) and a second analysis means (32), the second counting means (30) counting the number of points within the filtered cloud (34) of the filtered stream (16), said second analysis means (32) performing an analysis of the filtered cloud (34) when the number of points within said filtered cloud (34) is greater than a point threshold (PS).

10. Perception system (1) according to claim 9, wherein the second analysis means (32) is configured to deliver the filtered cloud (34) to the first organ (8) of the driving module (6) or to the second organ (10) of the driving module (6) according to the result of a comparison between an intensity of the points of said filtered cloud (34) and an intensity level (NI).

11. Perception system (1) according to claim 9, wherein when the number of points in the filtered cloud (34) is less than the point threshold (SP), the second counting means (30) delivers the filtered cloud (34) to the second organ (10) of the driving module (6).

12. A method for controlling a driving mode of a vehicle, said vehicle comprising a perception system (1) according to any one of claims 1 to 12, said method comprising: a first step in which the detection module (2) scans and models said environment in order to produce the first stream (3) of point clouds, a second step in which the first analysis device (12) of the adaptation module (4) analyzes the raw cloud (24) of the first stream (3) and then sends a signal (14) causing the filtering device (14) to switch between the active (ON) and inactive (OFF) modes, an optional third step in which, when the filtering device (14) is in active (ON) mode, the filtered cloud (34) of the filtered stream (16) is delivered according to information relating to the filtered cloud (34) to the first organ (8) or the second organ (10) of the driving module (6).

13. A method for controlling a driving mode of a vehicle according to claim 12 in combination with claims 4 to 6, wherein the second step comprises: a first substep in which the first counting means (20) counts the number of point clouds of the first stream (3) over a given period, a second substep in which, when the number of point clouds counted by the first counting means (20) reaches the cloud threshold (SN), the first analysis means (22) performs an analysis of the raw cloud (24) of the first stream (3) by comparing the intensity of points of said raw cloud (24) with the intensity threshold (SI) and then controls the transition of the filtering device (14) between the active mode (ON) and the inactive mode (OFF) according to the result of this analysis, a third substep in which the first counting means (20) is reset to zero.

14. A method for controlling a driving mode of a vehicle according to any one of claims 12 or 13 in combination with claims 9 to 11, wherein the optional third step comprises: a first sub-step in which the second counting means (30) counts the number of points within the filtered cloud (34) of the filtered flow (16), a second sub-step in which: when the number of points in the filtered cloud (34) is less than the point threshold (SP), the second counting means (30) delivers said filtered cloud (34) to the second organ (10) of the control module (6), when the number of points in said filtered cloud (34) is greater than the point threshold (SP), the second analysis means (32) performs an analysis in which the second analysis device (18) delivers the filtered cloud (34) to the first organ (8) of the control module (6) or to the second organ (10) of the control module (6) according to the result of a comparison between the intensity of the points of said filtered cloud (34) and the intensity level (NI).

Images