Early tracking of lateral objects

Abstract

The invention relates to a method for tracking a neighboring vehicle in the environment of a host vehicle, the method being implemented by a driver assistance device of the host vehicle, the driver assistance device being connected to a camera capable of capturing an image within a field of view and at a capture time, the method comprising: - detecting a first lateral object with respect to the host vehicle from a first image captured at a first capture time; - determining a data item associated with the first detected object; - creating a virtual object associated with a parameter, the parameter comprising a current position associated with the virtual object, wherein the current position is estimated from the data item; - estimating a future position of the virtual object from the current position; - updating the parameter from the future position with an update of a current length of the virtual object in the event that a first criterion is met.

Description

Technical Field

[0001] This disclosure relates to the field of driver assistance systems, and more specifically, to the automatic detection of vehicles in a region of interest (particularly the lateral region). Background Technology

[0002] The rise of Intelligent Transportation Systems (ITS) has spurred the development of numerous onboard systems for vehicles, particularly highway transport vehicles. These systems include, in particular, driver assistance systems and autonomous driving systems. Specifically, the detection and / or tracking of objects, including vehicles, plays a crucial role in issues such as traffic flow, road safety, and road infrastructure management (e.g., variable message signs and automatic speed cameras).

[0003] For example, in the context of onboard systems for autonomous or semi-autonomous vehicles operating in road traffic (e.g., on highways), early detection and tracking of objects around the vehicle in question are particularly important to prevent collisions between the vehicle in question and other vehicles. Specifically, in lane change assist or automatic lane change systems, early detection and accurate tracking of obstacles (such as the median strip or another vehicle in a lane adjacent to the vehicle in question) are required. In fact, if the autonomous vehicle in question fails to detect (or fails to detect in a stable manner or fails to detect sufficiently in advance) another vehicle in the adjacent lane and suddenly swerves, there is a risk of collision between the two vehicles.

[0004] In this context, most existing vehicle detection and tracking methods are based on classification algorithms that identify vehicles from image sequences or video streams captured by vision sensors (typically cameras). These vision sensors are usually located at the front and / or rear of the vehicle in question to acquire images of the area in front of and / or behind the vehicle. Field of view Images and / or image sequences. Then, the classification algorithm can identify one or more vehicles adjacent to the vehicle in question based on the frontal recognition of neighboring vehicles (i.e., to identify and classify the vehicle based on the front or rear of the vehicle detected in the field of view).

[0005] However, such detection and tracking methods using vehicle classifiers prove ineffective when the front or rear of an adjacent vehicle has not yet been detected within the field of view, even if such an adjacent vehicle is partially visible within the field of view, for example, via a partial lateral view (or one side). Therefore, if the vehicle in question is in a highway lane and an adjacent vehicle is relatively level (or slightly behind) it in the adjacent lane, a camera positioned behind the vehicle in question will capture the rear portion of that adjacent vehicle. In this case, the front of the adjacent vehicle is not yet visible to the camera positioned behind the vehicle in question, for example, until the speed difference between the vehicle in question and the adjacent vehicle is large enough for the vehicle in question to completely overtake the adjacent vehicle. Only when such a complete overtake occurs is the adjacent vehicle essentially at the same level as the vehicle in question, and the vehicle in question changing lanes into the adjacent lane will result in a collision with the adjacent vehicle. Such a critical situation cannot be detected by existing classification methods.

[0006] Furthermore, most existing vehicle tracking methods rely on predefined dimensions associated with the type of vehicle being detected. These dimensions are typically not updated, so even if vehicle tracking is performed (e.g., via a classifier), it cannot provide reliable information about the vehicle's size, and therefore does not allow for accurate tracking of the space occupied by the detected vehicle within the host vehicle's environment.

[0007] Therefore, it is necessary to ensure the safety of decision-making in driver assistance and / or autonomous driving systems, especially in the context of lane-change decisions. Specifically, it is necessary to detect, track, and perceive such surrounding vehicles early and reliably, even before the front or rear of vehicles around the target vehicle are visible within the field of view of the target vehicle's visual sensors—and therefore before it becomes possible to routinely track such surrounding vehicles using existing classifiers. Summary of the Invention

[0008] This disclosure helps to improve this situation.

[0009] A method is proposed for tracking at least one neighboring vehicle in the environment of a master vehicle, wherein both the neighboring vehicle and the master vehicle are motor vehicles. The method is implemented by a device configured to provide driver assistance functions for the master vehicle, the device being connected to at least one camera on the master vehicle and capable of acquiring images of the scene surrounding the master vehicle based on at least one field of view and at an acquisition time. The method includes the following steps: - Detect a first object having a vertical plane extending laterally relative to the main vehicle, based at least on the first image acquired at the first acquisition time. - Determine at least one data item associated with the first position of the detected first object based on a predefined coordinate system. - Create a virtual object associated with a set of parameters, which includes at least the current position associated with the virtual object, estimated based on at least data items related to a first position according to a predefined coordinate system. - Estimate the future position of the virtual object based at least on its current position, and - Update the parameters associated with the virtual object based at least on the future position, and this update of the parameters associated with the virtual object includes an update of the current length of the virtual object if the first criterion is satisfied.

[0010] Therefore, the proposed method allows the master vehicle to perform advance tracking of adjacent vehicles, specifically, once such an adjacent vehicle enters—even partially enters—the field of view of the camera on the master vehicle. In other words, the proposed method enables the tracking of vehicles traveling alongside the master vehicle even when the front or rear of the adjacent vehicle—which most existing classifiers rely on for vehicle tracking—is not visible.

[0011] Furthermore, the proposed method allows for updating the vehicle's length during the tracking process. Therefore, unlike most existing classifiers, the proposed tracking method does not rely solely on pre-existing vehicle dimensions but allows for updating the vehicle's dimensions.

[0012] Driver assistance features of a primary vehicle refer to the characteristics of a system on the primary vehicle that assists, guides, or even determines the primary vehicle's path. Such driver assistance features can be implemented in the context of semi-autonomous or autonomous vehicles, for example, with or without a driver. For instance, driver assistance features may include assistance with driving, lane changing, overtaking, or parking.

[0013] The field of view refers to the portion of the environment surrounding the main vehicle that is covered by the camera's field of view. When updating the vehicle's length, the field of view discussed is the rear field of view of the main vehicle.

[0014] Images acquired at the acquisition time refer to a sequence of continuous or discrete (i.e., two-dimensional) images of the environmental portion included in the field of view, associated with the time series in which these images were acquired. Specifically, the motion of elements around the main vehicle over time means that the field of view covers a constantly changing environment: the vehicle and elements adjacent to the main vehicle exhibit relative motion with respect to the main vehicle (as the main vehicle moves) and / or absolute motion within the environment.

[0015] The first object refers to an object detected within the field of view. Specifically, such an object can be associated with the motion (at least relative to the host vehicle) of a moving body (i.e., a non-inertial body) in the environment of an adjacent vehicle. For example, such a moving body could be another vehicle adjacent to the host vehicle, an obstacle in the environment, or even a bird. For example, such a first object can correspond to optical flow detected in an image, a set or segment of coplanar optical flow detected in an image, or a polygon or instance detected in an image (e.g., by a classifier). Specifically, a first object having a vertical plane extending laterally relative to the host vehicle refers to a detected object that is laterally positioned relative to the host vehicle. In other words, it is a so-called lateral view or corner view of the object detected within the field of view. For example, such a lateral side is detected when a vehicle is traveling in a lane adjacent to the host vehicle and is behind the host vehicle.

[0016] The data item associated with the first location of the first detected object refers to a measured and / or estimated or determined value that allows the detected object to be located within the acquired image and / or within the environment. For example, such a data item could represent the distance between the detected object and the host vehicle, or two-dimensional and / or three-dimensional coordinates. For example, the first location could refer to the position of a point belonging to the detected object (in the environment and / or in the image).

[0017] A virtual object is an object that has been virtually modeled or created and can be superimposed onto an image acquired by a camera. Such a created virtual object is also associated with the detected object because it is estimated to be superimposed on a location assumed to be occupied by a detected neighboring vehicle. Specifically, such a virtual object is characterized by its location in the environment and / or the image. Such a virtual object can also be characterized by its size, which reflects the estimated size of the detected vehicle. Therefore, virtual objects are created in this way to model the space occupied by neighboring bodies (e.g., neighboring vehicles) detected via the detected object.

[0018] The parameter set refers to parameter values ​​associated with and therefore with the space occupied by the virtual object. This parameter set can then include the current position of the virtual object, its current velocity, and / or its current size. This parameter set then allows the virtual object to be modeled in a reference frame of the image and / or environment. Specifically, the current position associated with the virtual object can refer to the position of a reference point associated with the virtual object; for example, such a reference point is modeled as belonging to the front side (and therefore the front end) of a detected adjacent vehicle. Such a current position may be imprecise, specifically because, at the stage of implementing the proposed method, the front side of the detected vehicle may not be visible in the field of view. Therefore, the current position can be determined based on one or more assumptions that specify such a current position within the image, for example, at the edge of the image.

[0019] A future position refers to a position estimated or predicted by a data estimator. Specifically, such an estimation or prediction of a future position can be based on at least current data fed into the estimator. In one embodiment, the future position of a virtual object is predicted and / or estimated by a Kalman filter based at least on the current position of the virtual object.

[0020] According to another aspect, an apparatus configured to provide driver assistance functions for a master vehicle is proposed. The apparatus is connected to at least one camera on the master vehicle and is capable of acquiring images of the scene around the master vehicle based on at least one field of view and at an acquisition time. The apparatus includes at least one processing circuit configured to implement a method for tracking at least one adjacent vehicle in the environment of the master vehicle.

[0021] According to another aspect, a computer program is proposed that includes instructions for implementing a tracing method when the program is executed by a processor.

[0022] According to another aspect, a non-transitory recording medium is provided that stores instructions for implementing a tracking method when executed by at least one processor via a program.

[0023] The features described in the following paragraphs can be implemented optionally, either independently or in combination: In one embodiment, updating the current length of a virtual object includes extending the length of the virtual object by a value determined from the interval between the current position and the future position, while keeping the current position of the virtual object constant.

[0024] Therefore, updating the parameters allows projecting the positional difference between the current and future positions, which are predicted based on at least one such current position, while taking into account the length of the virtual object. Specifically, the current position is maintained. This tracking method is particularly advantageous, especially when the front side of the detected neighboring vehicle is not yet visible in the field of view. In fact, maintaining the current position allows for an accurate representation of situations where the neighboring vehicle is not yet fully visible in the field of view (i.e., including its front side) when the current position of the virtual object is associated with, for example, the position of the front side of the virtual object (e.g., modeling the front side of the detected neighboring vehicle). Furthermore, once the front side of the vehicle is visible and detectable—for example, by a classifier—the proposed method effectively ensures continuous tracking of neighboring vehicles, where there is little change between the position of the virtual object detected by the classifier and the position of the virtual object detected using the proposed method. Moreover, by incorporating this positional difference, the length of the detected neighboring vehicle can be estimated, and thus the space occupied by the created virtual object can be updated. Therefore, the proposed tracking method allows for updating the vehicle's tracking parameters, which differ from the future position estimated, for example, by a data estimator.

[0025] In one embodiment, if a first comparison between the future position and the current position indicates the backward position of the virtual object relative to the current position in the acquired first image, then the current length of the virtual object is updated.

[0026] Therefore, the proposed tracking method allows updating the length of virtual objects modeled for adjacent vehicles based on an initial comparison of the virtual object's current and predicted positions. Specifically, a systematic elongation of the virtual object can be achieved when it is detected that the virtual object has moved backward in the image (e.g., causing the lateral component of the future position to be smaller than the lateral component of the current position). Specifically, such an update to the length of the virtual object can be based solely on current and future parameters associated with the virtual object, without requiring new object detection in the acquired image.

[0027] In one embodiment, the data associated with the first location includes a first lateral distance associated with the first detected object for a first acquisition time, the first lateral distance being determined from measurements of the first detected object in the first acquired image.

[0028] Therefore, the proposed tracking method enables the determination of, for example, the existing lateral distance between the detected neighboring object and the main vehicle, corresponding to an adjacent vehicle. For instance, a first position corresponding to the location of a specific point of the first detected object can specifically enable the derivation of information regarding the proximity of a detected neighboring vehicle in the environment associated with the main vehicle.

[0029] The first lateral distance refers to the lateral or angular distance between the element associated with the first detected object and the host vehicle. For example, in the case where the host vehicle and an adjacent vehicle are traveling in two parallel lanes (e.g., two lanes on a highway), the lateral distance corresponds to the lateral spacing that separates the lanes occupied by the host vehicle and the adjacent vehicle. Therefore, such a lateral distance reflects the interval between the host vehicle and adjacent vehicles in the environment.

[0030] In one embodiment, the future location is associated with the lateral distance of the object from the virtual object.

[0031] Therefore, once a virtual object has been created, its movement in the environment can be estimated, for example by a data estimator, making it possible to estimate predicted or future parameters for such a virtual object, such as the object's future position and lateral distance.

[0032] In one embodiment, the tracking method further includes: - Obtain a data item associated with a second location of a second detected object from at least one second image acquired at a second acquisition time following the first acquisition time. This data item associated with the second location includes a second lateral distance associated with the second detected object for the second acquisition time. - If the second detected object is determined to belong to a virtual object, a second lateral distance sum is compared in the second comparison, wherein the update of the parameters associated with the virtual object further depends on the result of the second comparison.

[0033] Therefore, the proposed tracking method allows for the consideration of positional changes of neighboring vehicles detected in the environment. Consequently, in a second image acquired at a later acquisition time than the first acquisition time, the second detected object may potentially differ from the first detected object, and this allows for updating the evolution of space occupied by neighboring vehicles in the environment (e.g., neighboring vehicles "increasingly" enter the field of view, or the velocity difference between the main vehicle and neighboring vehicles causing the main vehicle to have completely overtaken the neighboring vehicles and their front sides are now visible in the field of view). Therefore, the proposed update allows for updating virtual objects based on a second criterion, different from the first criterion.

[0034] The second detected object refers to a distinct object, regardless of whether it is separated from the first detected object in the acquired image. Such a second detected object can then correspond to an instance, polygon, optical flow, or segment reflecting the movement of adjacent vehicles detected in the environment. The second location associated with such a second detected object can then reflect the new space occupied by the adjacent vehicle.

[0035] In one embodiment, if the second lateral distance is strictly less than the lateral distance of the object, then updating the parameters associated with the virtual object includes updating the current position of the virtual object, which includes at least one update of the lateral component of the current position of the virtual object, which corresponds to the repositioning of the virtual object.

[0036] Therefore, the length update depends on a comparison of the lateral distances of continuously detected objects. Specifically, if the second lateral distance is greater than or equal to the estimated lateral distance of the virtual object, there is no need to update the virtual object's current position, as it is assumed that neighboring vehicles are gradually entering the field of view. Thus, the proposed tracking method utilizes the potential absence of the visible front side in the field of view to maintain the vehicle's current position (and thus avoids modeling virtual objects "moving backward" in the acquired image) and compensates for the difference between the current and future positions by laterally extending the length of neighboring vehicles.

[0037] Specifically, this method is advantageous because it corrects for a potential overestimation of the lateral distance determined and / or predicted between adjacent vehicles and the master vehicle. Therefore, such correction allows the tracking method to become more accurate and avoids dangerous situations where adjacent vehicles will actually be closer to each other than estimated (the second lateral distance corresponding to a more recently detected second object is shorter than the estimated lateral distance).

[0038] Therefore, the proposed tracking method allows for consideration of inaccuracies associated with creating virtual objects based on parameters and current position. In fact, if the second lateral distance is strictly less than the object's lateral distance, the virtual object estimated based on the object's lateral distance does not reflect the space actually occupied by neighboring vehicles; specifically, the proximity of neighboring vehicles to the main vehicle is underestimated. Therefore, the proposed method allows for correction of such potentially critical situations (since neighboring vehicles are potentially closer to the main vehicle than the position modeled by the virtual object based on its current position) by updating the virtual object's lateral position to reflect the actual lateral proximity between neighboring vehicles and the main vehicle. Thus, the method effectively guarantees consistency between the updates to the virtual object and the detectable content in the second image.

[0039] In one embodiment, if the second lateral distance is strictly less than the lateral distance of the object, updating the parameters associated with the virtual object includes updating the current position of the virtual object. This update of the current position further includes updating the height component of the current position of the virtual object, which corresponds to the repositioning of the virtual object. Such an update of the height component can also be based on terrain topology information in the environment.

[0040] In one embodiment, the current position of the virtual object is associated with the position of a first reference point belonging to the virtual object, the position of which is determined such that the first reference point is located on the vertical edge of the first image in a predefined coordinate system corresponding to the image reference system in the acquired image.

[0041] Therefore, even in the case of a partial side view of the vehicle thus detected, the current position allows us to model the positions of adjacent vehicles in the environment. The current position is then determined by constraints on modeling the positions of adjacent vehicles along the vertical edge of the image.

[0042] The first reference point is a point estimated to belong to a neighboring vehicle detected in the environment. Specifically, such a first reference point can be used to model the position of the front side of a neighboring vehicle (which may not yet be visible in the field of view).

[0043] In one embodiment, the virtual object is created based on the virtual object's position and a predefined size associated with a motor vehicle type, wherein at least one predefined size corresponds to the length of the virtual object.

[0044] Therefore, despite the lack of information about the size of the detected vehicles (at least during the initial iterations of the tracking method), the proposed tracking method enables the modeling of the space occupied by adjacent vehicles via virtual objects.

[0045] The length of a virtual object refers to its lateral length. Such virtual objects are used to model adjacent vehicles based on objects detected at angles relative to the main vehicle (based on the lateral plane).

[0046] In one embodiment, the future position of the virtual object is estimated based on a plurality of positions of the virtual object, which are determined separately for the acquisition time, until at least the front of at least one adjacent vehicle is detected in the field of view of the camera in at least one of the acquired images.

[0047] Therefore, the proposed tracking method can be advantageously implemented before the front (or frontal) side of adjacent vehicles becomes visible, enabling early tracking of vehicles adjacent to the main vehicle. Thus, when the front (or frontal) side of an adjacent vehicle enters the field of view, the existing classifier can take over vehicle tracking by updating the virtual object.

[0048] In one embodiment, each of the multiple locations of the virtual object is associated with an uncertainty value that is related to the uncertainty taking into account the initial conditions, and that the uncertainty value is greater than a predefined uncertainty threshold.

[0049] Therefore, the parameters determined to model the virtual object based on the angular view of detected neighboring vehicles can be assigned an uncertainty value greater than a threshold, which corresponds, for example, to the uncertainty value of data determined by a conventional classifier. Thus, when the parameters determined within the framework of the proposed method are fed into a data estimator to predict future positions, weights corresponding to the degree of error are assigned to these parameters. Such uncertainty values ​​can, for example, correspond to variance or covariance. Therefore, when the estimator predicting future positions receives additional parameter values ​​with lower uncertainty, it can predict the future positions of the virtual objects based on those values. For example, when the front of a neighboring vehicle enters the field of view, a vehicle front-end classifier can then provide the estimator with more reliable data than that obtained from positions estimated based on detected lateral objects. Therefore, the proposed method enables transformation between different data sources, thereby allowing the tracking of neighboring vehicles.

[0050] In one embodiment, the parameters associated with the virtual object are updated while keeping one parameter of the virtual object constant.

[0051] Therefore, the proposed method ensures that the virtual object remains stable when updating the parameters used to track neighboring vehicles. In fact, the proposed method allows updating both the length and / or position of the virtual object. Then, the parameters associated with the virtual object remain constant; for example, when the length of the virtual object is updated, at least one component of the position associated with the virtual object remains constant, or when the position of the virtual object is updated, the time parameters associated with the virtual object (such as collision time or the virtual object's velocity multiplied by distance) remain constant. Attached Figure Description

[0052] Other features, details, and advantages will become apparent after reading the following detailed description and analyzing the accompanying drawings, in which: [ Figure 1 ] Figure 1 A schematic representation of a main vehicle according to one embodiment is shown.

[0053] [ Figure 2 ] Figure 2 A bird's-eye view of the scene surrounding the main vehicle according to one embodiment is shown.

[0054] [ Figure 3 ] Figure 3 A schematic representation of a driver assistance device according to one embodiment is shown.

[0055] [ Figure 4 ] Figure 4 The steps of a vehicle tracking method according to one embodiment are shown.

[0056] [ Figure 5 ] Figure 5 A view of the scene around the master vehicle at a first acquisition time, according to one embodiment, is shown.

[0057] [ Figure 6 ] Figure 6 A virtual tracking object created according to one embodiment is shown.

[0058] [ Figure 7 ] Figure 7 An update of a created virtual tracking object is shown according to one embodiment.

[0059] [ Figure 8 ] Figure 8 A view of the scene around the master vehicle at a second acquisition time, according to one embodiment, is shown.

[0060] [ Figure 9 ] Figure 9 An update of a created virtual tracking object is shown according to one embodiment.

[0061] [ Figure 10 ] Figure 10 A view of the scene around the master vehicle at a third acquisition time is shown according to one embodiment.

[0062] [ Figure 11 ] Figure 11 The measurement performed on a view of the scene surrounding the master vehicle at a first acquisition time is shown according to one embodiment.

[0063] [ Figure 12 ] Figure 12 A reference point associated with an adjacent vehicle is shown according to one embodiment. Detailed Implementation

[0064] refer to Figure 1 . Figure 1 The diagram schematically illustrates the main vehicle (VP). The main vehicle (VP) can be a motor vehicle. There are no restrictions on the type of vehicle to which the VP belongs. The VP can be, for example, a private vehicle, a commercial vehicle, or an industrial vehicle, and can correspond to, for example, a car, van, two-wheeler, truck, or even a bus. The VP can also be a tractor vehicle, towing, for example, a trailer, semi-trailer, or caravan. The dimensions of the VP can be between 2 meters and 20 meters in length, between 0.5 meters and 5 meters in width, and between 1 meter and 5 meters in height.

[0065] The main vehicle VP is equipped with at least one onboard system that provides multiple functions or applications for the main vehicle VP. Such functions may include, for example, speed control, power steering, automatic airbag deployment, automatic headlight adjustment, etc.

[0066] The main vehicle VP is specifically equipped with an onboard system that enables the detection of objects around the main vehicle VP, for example as part of driver assistance functions such as obstacle detection, lane change assist, or automatic lane change. For this purpose, the onboard system of the main vehicle VP includes a driver assistance device 2. Device 2 is capable of providing driver assistance functions based on multiple views (or images) of the environment of the main vehicle VP. For this purpose, device 2 is connected to one or more vision sensors, such as an image capture device or camera 1 (in the remainder of the description, the vision sensor will be considered as camera 1). Device 2 is also capable of transmitting or conveying data, for example, related to the provided driver assistance functions. For this purpose, device 2 may be connected to a communication interface 50. In a particular embodiment, interface 50 may be integrated into device 2. For example, such a communication interface 50 may be a human-machine interface. Interface 50 may include a display screen, a touchscreen, a dashboard, and / or a speaker. The data transmitted from device 2 to interface 50 may, for example, correspond to assistance information instructing the driver of the main vehicle VP whether they may change lanes.

[0067] Now for reference Figure 2 . Figure 2 The environment (or scenario) where the main vehicle VP is located is shown in the ENV. Figure 2 This is an aerial view of the ENV scene.

[0068] An environment (ENV) can be defined in a three-dimensional reference frame (X,Y,Z) (referred to as the "world reference frame"), such as... Figure 1 and Figure 2 As shown, the origin of such a world reference frame (X,Y,Z) is predefined and fixed in the environment ENV.

[0069] The main vehicle (VP) is considered to be moving within the scene (ENV). For example, such a scene (ENV) corresponds to a road or highway consisting of several traffic lanes. These traffic lanes may be obviously parallel to each other, such as... Figure 2 The dotted vertical line in the image is shown. For example, Figure 2 Three traffic lanes are shown, with the main vehicle VP located in the middle lane. The main vehicle VP is considered to be moving along the principal direction X of the reference frame (X,Y,Z), which is called the longitudinal direction. Such motion is... Figure 1 and Figure 2The arrows attached to the main vehicle VP are schematically shown in the diagram. The main vehicle VP may also have movement in the Y direction, known as the lateral direction, along the reference frame (X,Y,Z). For example, such movement in the Y direction, referred to as lateral movement (or displacement), can occur when the main vehicle (VP) changes traffic lanes. Displacement of the main vehicle VP in the Z direction (i.e., in height) is not considered. Therefore, the main vehicle VP is considered to be maintained on the ground and thus has a height in the Z direction that roughly corresponds to a predefined dimension of the main vehicle VP. In one embodiment, such height in the Z direction of the main vehicle VP may vary on the order of centimeters or decimeters, and such height variation in the Z direction may be linked, for example, to undulations or roughness in the main vehicle VP's shock absorbers and / or the environmental ENV (specifically, in the traffic lane of the main vehicle VP). In the following description, such height in the Z direction of the main vehicle VP is considered to be known.

[0070] The scene ENV surrounding the main vehicle VP may also include other element OVs and vehicles VV1, VV2, and VV3. Vehicles VV1, VV2, and VV3 are adjacent vehicles of the main vehicle VP. These adjacent vehicles VV1, VV2, and VV3, like the main vehicle VP, are moving motor vehicles within the scene ENV. These adjacent vehicles VV1, VV2, and VV3 may specifically have similar or different dimensions and (e.g., in terms of speed or acceleration) movement characteristics from the main vehicle VP. For example, Figure 2 An object can represent a primary vehicle (VP) traveling in a highway lane and adjacent vehicles (VV1, VV2, VV3) traveling in a highway lane adjacent to the lane used by the primary vehicle (VP). An object (OV) is an element adjacent to the primary vehicle (VP) and can refer to any different element of the adjacent vehicles (VV1, VV2, VV3). For example, an adjacent OV element can correspond to an obstacle in the scene, such as a median separating two traffic lanes, a road sign, or even a bird flying in the scene's ENV.

[0071] like Figure 2 As shown, the main vehicle VP is believed to be equipped with a camera 1 positioned at the rear of the main vehicle VP. In another embodiment ( Figure 2 (Not shown in the image), camera 1 can be positioned at the front of the main vehicle VP, or multiple cameras 1 can be positioned at both the front and rear of the main vehicle VP. In the context of this description, camera 1 is considered to be positioned at the rear of the main vehicle VP, and the field of view (FOV) is the rearward field of view of the main vehicle VP. Camera 1 allows for the capture of an image (or view) of the scene ENV of the main vehicle VP according to the FOV. Such a FOV depends specifically on the type of camera 1 used and the positioning of camera 1 within (or on) the main vehicle VP. Reference Figure 2The field of view (FOV) covers a portion of the scene's active field of view (ENV), thus capturing a limited portion of the scene's ENV based on the FOV. Therefore, in Figure 2 In the diagram, the field of view (FOV) covers a portion of the lane where the main vehicle (VP) is located, as well as corresponding portions of adjacent lanes. Specifically, the FOV covers the portion of the scene ENV where the adjacent vehicle (VV2) is completely contained, and the portion of the scene ENV where the adjacent vehicle (VV1) is partially contained (the unfilled gray area). For example, as shown... Figure 2 As shown, the left rear end of adjacent vehicle VV1 (e.g., including the left rear wheel of adjacent vehicle VV2) falls within the field of view (FOV). However, reference point W, for example, located in the middle of the front side of the vehicle, is not within the FOV. Figure 2 Within the field of view (FOV) of the host vehicle (VP). Adjacent vehicle VV3 is not visible within the FOV of the host vehicle (VP). A portion of adjacent vehicle VV1 is not visible within the FOV of the host vehicle (VP). Figure 2 The striped areas represent portions of the view from adjacent vehicles VV1 and VV3 that are not visible in the field of view (FOV). Adjacent objects (OV) (e.g., birds) are visible in the FOV.

[0072] In the scenario ENV (such as Figure 2 In the context of the scenario shown in the reference frame (X,Y,Z), assuming, for example, the main vehicle VP has a known main velocity... Driving. To simplify the rest of the description, this main speed... V is considered constant VP Standard. Adjacent vehicle VV1 is considered to have an unknown adjacent speed relative to the driver assistance device 2 of the primary vehicle VP. Driving. In the embodiments described below, it can be assumed that adjacent speeds are used. V VP The standard is lower than that used for main speed V VP Standard. Under this assumption, and assuming the adjacent speeds of adjacent vehicles VV1. It remains substantially constant over a given time interval, just like the main speed of the main vehicle VP. Similarly, the distance between the master vehicle VP and the adjacent vehicle VV1 along the longitudinal direction X will increase over time, such that at a future time, the master vehicle VP will exceed the adjacent vehicle VV1, and the adjacent vehicle VV1 will be completely within the (rear) field of view (FOV) of the master vehicle VP, just as the adjacent vehicle VV2 was initially.

[0073] Figure 5 , Figure 8 and Figure 10 The changes in the position of the adjacent vehicle VV1 within the field of view (FOV) of the main vehicle VP over time are shown.

[0074] refer to Figure 5 , Figure 8 and Figure 10 . Figure 5 , Figure 8 and Figure 10 This schematically illustrates how... Figure 2 The image shows the view of the scene ENV surrounding the main vehicle VP as seen from the (rear) field of view (FOV) of camera 1, positioned as the main vehicle VP. To improve... Figure 5 , Figure 8 and Figure 10 Clarity Figure 5 , Figure 8 and Figure 10 The image only shows a portion of the adjacent vehicle VV1 visible within the field of view (FOV); Figure 5 , Figure 8 and Figure 10 The view does not show the field of view (FOV) including the data from... Figure 2 The adjacent vehicle VV2 and the adjacent element OV.

[0075] Figure 5 , Figure 8 and Figure 10 This can correspond to images acquired consecutively by camera 1 at consecutive times (or acquisition times) T1, T2, and T3. Each of the times T1, T2, and T3 can be, for example, time-intervals of one millisecond or more. Each image corresponds to a set of pixels with definable coordinates in a two-dimensional reference frame (y, z), which is called the "image reference frame," such as... Figure 5 , 8 As shown in Figure 10. For example, the lower left pixel of each image captured by camera 1 can correspond to the origin of the image reference frame (y, z). In other embodiments, the center pixel of the image captured by camera 1, the optical center of the image, the pixel in the image corresponding to the position of camera 1 in the environment ENV, or the pixel corresponding to the vanishing point C' in the captured image can be regarded as the origin of the reference frame (y, z). The resolution of the captured images (and therefore the number of pixels) is the same for all captured images and depends specifically on the specifications of camera 1.

[0076] Figure 5 The image is schematically represented by camera 1 at time T1, for example, with... Figure 2 The scene configurations shown correspond to the ENV and FOV configurations. Therefore, only the first, so-called lateral portion of the adjacent vehicle VV1 is visible. Figure 5 As can be seen in the image; the first lateral portion specifically includes the left rear wheel of the adjacent vehicle VV1.

[0077] Figure 8The image is schematically represented by camera 1 at time T2, following time T1. At such time T2, the second lateral portion of the adjacent vehicle VV1, which is larger than the first lateral portion, is... Figure 8 As can be seen in the image, such a second lateral portion always includes the left rear wheel and part of the left front wheel of the adjacent vehicle VV1. In other words, between time T1 and T2, the speed difference between the main vehicle VP and the adjacent vehicle VV1 causes the adjacent vehicle VV1 to "increasingly" enter the field of view (FOV) of camera 1 between time T1 and T2.

[0078] Figure 10 The image is schematically represented by camera 1 at time T3, following time T2. At such time T3, the complete adjacent vehicles VV1 are... Figure 10 This is visible in the image. Specifically, Figure 2 The reference point W, shown and located in front of the adjacent vehicle VV1, becomes visible in the acquired image; this reference point W is... Figure 10 The pixel w in the image shown corresponds to this. In other words, between time T2 and T3, the speed difference between the main vehicle VP and the adjacent vehicle VV1 causes the adjacent vehicle VV1 to "fully enter" the field of view (FOV) of camera 1 between time T2 and T3.

[0079] In the context of a driver assistance function provided by device 2, intended for example, assisting the primary vehicle VP in lane-changing maneuvers, a process is needed to detect and track vehicles surrounding the primary vehicle VP, such that if the primary vehicle VP risks colliding with one of the surrounding vehicles, the primary vehicle VP does not move into the adjacent traffic lane. Existing techniques for detecting and tracking vehicles surrounding the primary vehicle VP specifically rely on the use of a classifier, which is based on, for example... Convolutional Neural Network Methods (or CNN) K-Nearest Neighbors Algorithm (or KNN) or Support Vector Machine (or SVM). Such classifiers, capable of detecting and tracking vehicles around a master vehicle (VP), are specifically based on the detection and classification of frontal views (front or rear) of vehicles surrounding the master vehicle (VP). The use of such classifiers can specifically involve learning stages based on multiple images representing frontal views of various vehicle types. Therefore, refer to... Figure 2 This refers to the frontal view of the adjacent vehicle VV2 belonging to the field of view (FOV) of camera 1, which can be used to detect adjacent vehicle VV2 based on existing classification techniques. This also applies to... Figure 10 The adjacent vehicle VV1 in the image at time T3.

[0080] However, Figure 2 as well as Figure 5 and Figure 8 The positions of adjacent vehicles VV1 shown correspond to the following situations: due to Figure 5and Figure 8 The image shown does not contain a visible front view of the adjacent vehicle VV1. Figure 2 The second reference point W, located in front of the adjacent vehicle VV1, is not yet visible in the field of view (FOV) of camera 1 during these stages. Existing classifiers cannot, or at least cannot, effectively detect or track the adjacent vehicle VV1 without incurring significant cost and / or computation time. However, these situations correspond to the following critical scenario: in the absence of lane change assist information indicating that the adjacent vehicle VV1 has been detected near such a location, if the primary vehicle VP changes position and moves into the lane of the adjacent vehicle VV1, the primary vehicle VP may collide with the adjacent vehicle VV1.

[0081] Then, it was proposed and Figure 4 The document specifically describes a method for early tracking of neighboring vehicles VV1, specifically for detecting and tracking neighboring vehicles VV1. Figure 2 , Figure 5 and Figure 8 In the described scenario, this phase is referred to as "advance" because the front side of the tracked vehicle may not yet be visible within the field of view (FOV). Such a method for advance tracking of adjacent vehicles (VV1) is described by... Figure 3 The driver assistance system shown is implemented.

[0082] Now for reference Figure 3 . Figure 3 A schematic representation of the onboard system of the primary vehicle (VP) is shown. Specifically, such an onboard system corresponds to the driver assistance system of the primary vehicle (VP). When the primary vehicle (VP) is traveling in a traffic lane, the driver assistance system of the primary vehicle (VP) provides lane change assistance or automatic lane change functions, for example... Figure 2 As shown.

[0083] First, the system includes a driver assistance device 2. Device 2 itself includes: a unit 20 for object detection of acquired images; a unit 30 for determining parameters associated with the objects detected by unit 20; a unit 40 for creating virtual objects associated with the detected objects; a unit 50 for estimating data (such as a Kalman filter) for tracking the created virtual objects, for example through data fusion, data learning, and / or data prediction; and a unit 60 for updating the parameters associated with the created virtual objects.

[0084] The driver assistance device 2 is also connected to a visual sensor, such as a camera or video camera 1. Device 2 includes an input unit ( Figure 3(Not shown in the image), which allows device 2 to receive a data stream from camera 1 substantially in real time. Depending on the field of view (FOV) of camera 1, such a data stream corresponds to a discrete or continuous sequence of images (or views) of the scene's envelopment (ENV). Each received image is associated with the image acquisition time of camera 1. Each image may be timestamped.

[0085] On the other hand, the driver assistance device 2 can be connected to the communication interface 70. Such an interface 70 can correspond to a human-machine interface integrated into the onboard system of the main vehicle VP. Such an interface 70 can also be integrated into the device 2. The communication interface 70 can also be a remote interface. The communication interface 70 can include, for example, a display screen, touchscreen, dashboard, or speaker, which allows the transmission of driver assistance-related instructions for the main vehicle VP via, for example, visual, tactile, and / or auditory information. Specifically, the communication interface can enable the transmission of detection status of vehicles adjacent to the main vehicle VP or instructions related to possible lane changes by the main vehicle VP. Such information transmitted by the interface 70 can, for example, correspond to an aerial visual representation of the real-time estimated position of the main vehicle VP and the corresponding positions of adjacent elements belonging to the field of view (FOV) (e.g., as shown in the image). Figure 2 (As shown) or an audio stimulus or virtual object OV1 that warns of collision risk, or a highlighted and real-time updated overlay of the image stream from camera 1, for example Figure 6 , Figure 7 and Figure 9 As shown.

[0086] Units 20, 30, 40, 50, and 60 of device 2 each include processing circuitry, which includes at least one processor (21, 31, 41, 51, 61) and a memory unit (22, 32, 42, 52, 62) to implement one or more steps of a method for pre-tracking adjacent vehicles VV1, wherein the one or more steps will be... Figure 4 The description is as follows. Specifically, each processing unit 20, 30, 40, 50, and 60 of device 2 can rely on data processed and / or obtained by other units to implement one or more steps of the method for pre-tracking adjacent vehicles VV1, such as... Figure 3 As shown by the arrow in the image.

[0087] Now for reference Figure 4 . Figure 4 The sequence of steps for implementing a method for advance tracking of adjacent vehicles VV1 by a system including a driver assistance device 2 of the main vehicle VP is shown, as follows: Figure 3 shown. Specifically, Figure 4 The method described in the document for early tracking of neighboring vehicles VV1 includes a phase of detecting objects associated with neighboring vehicles VV1 and a phase of tracking the neighboring vehicles VV1 themselves based on the detected objects.

[0088] In step 400, device 2 receives from camera 1 multiple images (or views) associated with corresponding image acquisition times. Specifically, such images may be received continuously, for example, via a video stream. In this case, at step 400, device 2 may discretize the received video stream to obtain a discrete set of timestamped images associated with corresponding acquisition times.

[0089] Steps 410 to 450 described below are performed based on objects detected in the images acquired at a given acquisition time—for example, the first detected object OBJ1.

[0090] In step 410, the first object OBJ1 is detected on at least one of the acquired images. (Reference) Figure 5 The first detected object OBJ1 is represented on the image corresponding to the first acquisition time T1. In one embodiment, the first detected object OBJ1 is a so-called lateral object in the acquired image because the first detected object OBJ1 belongs to the lateral plane of the image, such a lateral plane being, for example, parallel to the plane (X,Z) in the world reference frame (X,Y,Z). The detection of such a first lateral object OBJ1 is then allegedly advanced because it occurs before the front of the adjacent vehicle VV1 becomes visible. Specifically, in step 410, the first detected object OBJ1 corresponds to a non-stationary body. In other words, the first detected object OBJ1 corresponds to a moving body in the environment ENV (and therefore in the world reference frame (X,Y,Z)) having a velocity (e.g., a so-called longitudinal velocity corresponding to displacement along the X-axis and / or a so-called lateral velocity and / or rotational velocity corresponding to displacement along the Y-axis in the world reference frame (X,Y,Z)).

[0091] In step 410, the first image (e.g.,) can be processed by processing one or more consecutively acquired images. Figure 5The first object OBJ1 is detected on the image. In one embodiment, the first object OBJ1 can be detected, for example, by a classifier capable of detecting lateral objects OBJ1 in the acquired image, such as a classifier for detecting the lateral portion of a vehicle, like a wheel detector. Alternatively, homography can be used to detect the first object OBJ1 on at least two consecutively acquired images to identify one or more connected sets or optical flows across these images at a relatively uniform speed within a specified time interval. According to this embodiment, detecting the first object OBJ1 involves determining a pixel displacement or optical flow with a substantially constant speed from one frame to the next. Each optical flow can be determined by matching points across several consecutive images, for example using the Lucas-Kanade method. Specifically, the determined optical flow can be divided into one or more segments that share a common display point or vanishing line and belong to a lateral plane within the FOV field of view. In other words, the acquired image can be segmented into one or more segments, where each segment defines a group of optical flows (and thus a group of pixels with substantially the same speed and describing the same motion within a given time interval).

[0092] Therefore, in step 410, the first detected object OBJ1 is a set of pixels, which can be composed of instances, contours, polygons, or rectangles (or "...") detected and classified by the object segmentation classifier in the acquired image. bounding box " ) Formation. Alternatively, the first detected object OBJ1 can be a set of pixels formed by optical flow or segments comprising several optical flows. Specifically, the set of pixels forming the first detected object OBJ1 can be a set of coplanar pixels parallel to the lateral plane in the world reference frame (X,Y,Z).

[0093] In step 420, the device 2 may determine data related to the first position of the first detected object OBJ1. Specifically, such data related to the first position of the first detected object OBJ1 may be a first lateral distance d associated with the first detected object OBJ1. lat,1 Specifically, such a first lateral distance d can be determined from values, measurements, estimates, or data associated with the first detected object OBJ1 in the first acquired image. lat,1 In one embodiment, such a first lateral distance d lat,1 This can be, for example, a roughly fixed value considering the region where the first detected object OBJ1 is located in the acquired image. In one embodiment, the first lateral distance dl can be determined based on multiple lines on the ground. at,1These multiple lines are visible on a first image acquired between the central vertical line VL (e.g., a dashed vertical line) and the line at the height of the first detected object OBJ1 on the ground, for example, in the case of a vehicle in a highway lane. In one embodiment, the first lateral distance d can be determined from data associated with the first detected object OBJ1 and / or data measured on the first acquired image. lat,1 Such data includes, for example, the size (in pixels) of the first detected object OBJ1 and / or the angle between pixels belonging to the first detected object OBJ1 and predefined reference pixels on the first acquired image. This data can be estimated by a classifier that has detected the first object OBJ1. It can also be estimated from measurements performed on the acquired first image. For example, a first lateral distance d can be determined from an estimate of a first position associated with the detected object OBJ1. lat,1 For example, such a first position could be the position of the first reference pixel b1 belonging to the detected object OBJ1. Figure 5 The image shows a first reference pixel b1. In one embodiment, such a first reference pixel b1 can be identified by an object classifier, for example, as a vehicle wheel. Alternatively, the first reference pixel b1 can be identified by measuring the minimum angle θmin, as shown in the image. Figure 11 As shown. To measure such a minimum angle θmin, consider the vanishing point C' of the first acquired image; this vanishing point is related to... Figure 11 The optical center of the first acquired image shown coincides with the vertical axis VL passing through the vanishing point C'. A first reference pixel b1 can then be selected from the set of pixels forming the detected object OBJ1 such that the angle θmin formed between the vertical axis VL and the axis connecting the vanishing point C' and such reference pixel b1 is minimized.

[0094] Then, the first reference pixel b1 obtained in this way has a first position in the first acquired image, which can be determined by the coordinates of point b1 in the image reference frame (y, z). Based on such coordinates in the image reference frame (y, z), the initial position associated with the detected object OBJ1 can be estimated in the world reference frame (X, Y, Z). Specifically, using pinhole modelThe coordinates of the first location in the world reference frame (X, Y, Z) can be determined based on the coordinates of the first reference pixel b1 in the image reference frame (y, z) and predefined characteristics of camera 1. Applying the pinhole model to obtain the coordinates associated with the first detected object OBJ1 in the world reference frame (X, Y, Z) depends specifically on several computational assumptions related to the properties of camera 1 and the environmental ENV associated with the world reference frame (X, Y, Z). For this purpose, classical extrinsic parameters of camera 1 can be considered. A "fisheye" type camera 1 can be used. Alternatively, the flat world assumption can be considered. Alternatively, georeferencing the environmental ENV via ground markings or other topographic measurements can also be considered. In the context of the applied pinhole model, any final geometric distortion that may be caused by the optical system of camera 1 can be ignored. In another embodiment, determining the coordinates associated with the first detected object OBJ1 in the world reference frame (X, Y, Z) may include correcting for lens distortion of camera 1. Then, based on the world reference frame (X, Y, Z), the first lateral distance d associated with the first detected object OBJ1 (real, in meters) can be directly derived from the coordinates along the Y-axis of the first position. lat,1 Specifically, the first lateral distance d associated with the first detected object OBJ1 can be obtained by calculating the difference between the coordinates of the first position along the Y-axis in the world reference frame (X, Y, Z) and the coordinates of the origin of the camera 1 on the main vehicle VP along the Y-axis. lat,1 For example, the origin coincides with the vanishing point C', the optical center of the image, or a point on the image associated with the position of camera 1 in the environment ENV. Alternatively, a combination of the above methods can be implemented in step 420 to determine the first lateral distance d associated with the first detected object OBJ1. lat,1 .

[0095] At the end of step 420, the first detected object OBJ1 on the first image acquired at the first acquisition time T1 can then be associated with a data item related to the first position of the first object OBJ1. Specifically, such a data item could be the first lateral distance d. lat,1 Such a first lateral distance d lat,1 Specifically, this corresponds to the estimation of the lateral distance separating the master vehicle VP from the detected object OBJ1. In other words, step 420 enables the estimation of the true lateral distance (i.e., in the environment ENV) between the master vehicle VP and the adjacent moving body corresponding to the first detected object OBJ1 (such an adjacent body is, for example, the adjacent vehicle VV1 of the master vehicle VP). In the following description, such an adjacent moving body corresponding to the first detected object OBJ1 is considered to be the adjacent vehicle VV1.

[0096] In step 430, the device 2 may determine and / or estimate a set of parameters associated with the thus detected neighboring vehicle VV1. These parameters may specifically be derived from the first lateral distance d associated with the first detected object OBJ1. lat,1 This is used to estimate and correlate with neighboring vehicle VV1. For example, such parameters could include the current position associated with the detected neighboring vehicle VV1. The current speed associated with the detected adjacent vehicle VV1 And / or the current size associated with the detected adjacent vehicle VV1 .

[0097] Therefore, in step 430, determining the set of parameters associated with the neighboring vehicle VV1 may include determining the current speed associated with the neighboring vehicle VV1. In one embodiment, such a current speed The rotational speed can be estimated, for example, based on observations of movement of points located on the wheel by a wheel classifier. Alternatively, for example, it can be estimated from the first lateral distance d associated with the first detected object OBJ1, as determined in step 420. lat,1 And derive the current velocity from the time data item iTTC associated with the first detected object OBJ1. Such a time data item iTTC can be correlated with the reciprocal of the collision time associated with the first detected object OBJ1. In the case of the first detected object OBJ1 corresponding to a segment (i.e., an optical flow set), such a time data item iTTC associated with each segment can be determined, for example, by one of the methods described in documents WO2018059629, WO2018059631, and WO2018059632.

[0098] Therefore, in step 430, the current speed associated with the adjacent vehicle VV1 The velocity that can be identified as associated with the first detected object OBJ1 is: in: - It is the relative speed (in meters per second) associated with the first detected object OBJ1 (with respect to the master vehicle VP). - It is the lateral distance (in meters) associated with the first detected object OBJ1, and - iTTC is a time data item (represented as a time reciprocal) associated with the first detected object OBJ1.

[0099] Furthermore, in step 430, determining the set of parameters associated with the neighboring vehicle VV1 may include determining the current size associated with the neighboring vehicle VV1. In one embodiment, such current dimensions can be predefined based on previously existing dimensions typically found in vehicles. For example, previously existing dimensions can correspond to predefined length L1, predefined width L2, and predefined height H.

[0100] Finally, in step 430, determining the set of parameters associated with the neighboring vehicle VV1 may include determining the current position associated with the neighboring vehicle VV1. Specifically, this current position This can be associated with an estimated point on the adjacent vehicle VV1 (e.g., a first reference point W1 associated with the adjacent vehicle VV1). Specifically, this first reference point W1 can be associated with the center of the front side of the adjacent vehicle VV1 (where it is located at...). Figure 5 (Not visible in the acquired image). Therefore, the current position associated with the neighboring vehicle VV1 is estimated. This can be simplified to estimating the coordinates (X_W1, Y_W1, Z_W1) of the first reference point W1 in the world reference frame (X, Y, Z). In other words, in one embodiment: To estimate the coordinates (X_W1, Y_W1, Z_W1) of the first reference point W1, in the... Figure 12 In the illustrated embodiment, device 2 can use the coordinates (X_C, Y_C) of the center point C of camera 1 in a world reference frame (X, Y, Z), where such a center point C corresponds, for example, to an optical center or a point associated with the position of camera 1 or its sensor in the environment ENV (the projection of such a center point C onto the world reference frame is shown in the figure). Figure 5 (Point C' is generated). Device 2 can also rely particularly on the previously existing dimensions L1, L, and H of the adjacent vehicle VV1, the first lateral distance determined in step 420. The focal length of camera 1 And the sensor size sz_capt. Then, device 2 can be determined by applying the pinhole model and Thales' theorem: in: - ( () represents the coordinates of the first reference point W1 in the world reference frame (X,Y,Z). - ( , (x, y, z) represents the coordinates of the center point C of camera 1 in the world reference frame (X, Y, Z). - It is the first lateral distance associated with the first detected object OBJ1. - It is the component of the focal length of camera 1 along the X-axis. - It is the actual size of the sensor, and - It is the predefined width of the adjacent vehicle VV1.

[0101] Specifically, the coordinates of the first reference point W1 in the world reference frame (X,Y,Z) can be determined by assuming that the center point C coincides with the center of the acquired image. Optical distortion is negligible. Furthermore, the focal length... It is assumed that the coordinates of the first reference point W1 are the same on both the vertical (VL) and horizontal axes of the image. Specifically, the above formula describing the coordinates of the first reference point W1 depends on the relative positioning of the adjacent vehicle VV1 with respect to the master vehicle VP.

[0102] Finally, the coordinates Z_W of the first reference point W1 along the Z-axis in the world reference frame (X,Y,Z) can be determined to correspond to half the predefined height of the adjacent vehicle VV1, i.e., Z_W = H / 2. Alternatively, the coordinates Z_W can correspond to the coordinates along the Z-axis of the first position associated with the first detected object OBJ1. Specifically, the coordinates Z_W can correspond to the height of the point closest to the ground in the world reference frame (X,Y,Z), whose projection in the image reference frame (y,z) belongs to the first detected object OBJ1.

[0103] Therefore, at the end of step 430, device 2 can obtain the (current) parameters associated with the adjacent vehicle VV1 (and thus with the first detected object OBJ1). , , A set of parameters. , , This is associated with a given acquisition time—that is, in this case, the first acquisition time T1—which corresponds to a certain position and certain kinematic conditions of an adjacent vehicle VV1 in the environment ENV, which are estimated based on the first detected object OBJ1 at that first acquisition time T1.

[0104] Optionally, at the end of step 430, a step ( Figure 4 (Not shown) may include verifying the detection of adjacent vehicle VV1 from the detected object OBJ1 and the parameters determined from the first detected object OBJ1. , , The reasonableness of ). Then, a reasonableness check can optionally be implemented so as to, for example: - Verify current speed This reflects that adjacent vehicles move in a manner reasonably similar to that of adjacent vehicle VV1 and / or move in a direction closer to the master vehicle VP. - Verify that the height of the pixel set forming the first detected object OBJ1 is reasonable for the adjacent vehicle VV1 (e.g., the first detected object OBJ1 corresponding to the movement of a bird captured in the field of view FOV will potentially have a height that does not fall within the range of reasonable height values ​​determined to correspond to the vehicle).

[0105] In step 440, device 2 can verify whether the adjacent vehicle VV1 and / or the first detected object OBJ1 correspond to a vehicle previously detected by device 2, in order to associate the first detected object with the tracked object. Generally, at any given acquisition time, the parameters associated with the detected vehicle, as determined in step 430, enable device 2 to estimate the space occupied by the detected vehicle in the environmental ENV. For example, such space occupied by the detected vehicle can be estimated based on (e.g., estimated by a conventional classifier or determined in step 430) the vehicle's position and the size associated with that vehicle (e.g., a previously existing size). For example, such occupied space can be associated with a set of coordinates in the environmental ENV associated with the detected vehicle according to a world reference frame (X,Y,Z). For example, reference... Figure 2 The adjacent vehicle VV2 has already been detected by device 2 at the first acquisition time T1, for example by a conventional classifier that detects and tracks the front of the vehicle. Then, the occupied space corresponding to the set of locations in the ENV environment has been stored by device 2 in association with the adjacent vehicle VV2 during the first acquisition time T1. Therefore, at the first acquisition time T1, the location associated with the object detected at the first acquisition time T1 (e.g., the first location associated with the first detected object OBJ1 in step 410) can then be compared with the occupied space associated with the adjacent vehicle VV2 to verify whether the object detected at the first acquisition time T1 corresponds to the movement of the adjacent vehicle VV2.

[0106] Therefore, in step 440, the detection of an object can be verified by comparing the first position associated with the first detected object OBJ1 and / or the set of parameters estimated based on the first detected object OBJ1 and associated with the adjacent vehicle VV1 with one or more occupancy spaces (i.e., one or more sets of coordinates) associated with one or more bodies (e.g., vehicles) previously detected and stored by the device 2.

[0107] If, in step 440, the first position associated with the first detected object OBJ1 and / or the current position of the adjacent vehicle VV1 falls within the occupied space stored by device 2, then device 2 determines that the first detected object OBJ1 (and therefore the adjacent vehicle VV1) corresponds to an object of interest (e.g., a tracked vehicle) known to the device, which was detected before the first acquisition time T1. Therefore, device 2 does not need to create a new virtual object associated with the first detected object OBJ1 and can rely on data already associated with a known object of interest. Steps 400 to 440 allow tracking of such objects of interest (e.g., adjacent vehicles VV1) associated with the first detected object OBJ1. Tracking objects of interest includes estimating future parameters associated with such objects of interest (e.g., future position, future size, future speed, etc.) and updating the objects of interest. For this purpose, at step 450, the parameters associated with the adjacent vehicle VV1 ( , , ) and the first acquisition time T1 determined from the first detected object OBJ1 is fed to the Kalman filter (e.g., " Extended Kalman Filter "or EKF". Specifically, such a Kalman filter is configured based on at least the current parameters provided by device 2 ( , , To estimate the future (or predicted) parameters of the object of interest. , , ).

[0108] Once the object of interest has been detected and tracked, a Kalman filter or any other data estimator can estimate the future parameters associated with that object of interest. For example, at the end of step 450, the parameters associated with the neighboring vehicle VV1 ( , , The feed to the Kalman filter enables the tracking of adjacent vehicles VV1 based on the Kalman filter's continuous predictions of the position and velocity of adjacent vehicles VV1 in the environmental ENV. Specifically, the lateral distance to the object (corresponding to the varying lateral spacing between the master vehicle VP and the followed adjacent vehicle VV1) can be predicted. Therefore, depending on the current moment, any object of interest tracked by device 2 can be associated with a given lateral distance to the object and with a predicted set of parameters that allows estimation of the position and movement of the tracked object of interest relative to the master vehicle VP as the object of interest travels through the environmental ENV.

[0109] In step 470, based on the estimated future parameters of the tracked vehicles, the device can then continue to update the parameters associated with the tracked vehicles corresponding to the occupied space. This step 470, which updates the parameters associated with the detected vehicles, will be described in detail later in the specification.

[0110] If, in step 440, the first location associated with the first detected object OBJ1 and / or the current location of the adjacent vehicle VV1 is not included in the occupancy space stored by device 2, device 2 may proceed to step 460 to create a virtual object OV1 associated with the first detected object OBJ1 (and therefore with the detected adjacent vehicle VV1). Figure 6 An example of a virtual object OV1 created from the first detected object OBJ1 is shown. Specifically, the parameters determined in step 430 can be used ( , To create a virtual object OV1, for example, based on the coordinates (X_W1, Y_W1, Z_W1) determined from the first reference point W1 and using the model of camera 1, a first virtual reference point w1 with coordinates (yw1, zw1) in the image reference system (y,z) can be placed on the acquired image, as shown below. Figure 6 As shown. Specifically, a virtual point w1 can be located in the image reference frame (y,z) under the following assumptions: - The first detected object OBJ1 corresponds to the rear of the adjacent vehicle VV1 (e.g., it includes the rear wheels of the adjacent vehicle VV1). - The virtual reference point w1 is considered to be located on the lateral edge of the acquired image. In other words, the lateral portion visible on the acquired image is considered to represent the entire length of the adjacent vehicle VV1 in the state considered on the acquired image, and the first reference point W1 is considered to be located on the front side of the adjacent vehicle VV1.

[0111] In another embodiment, the first virtual reference point w1 may be located at a location other than the edge of the image, for example, inside or even outside the image.

[0112] Therefore, refer to Figure 6 If the pixel located at the lower left of the acquired image is considered the origin of the image reference frame (y,z), and the adjacent vehicle VV1 is detected as being on the right lateral side of the master vehicle VP (e.g., ... Figure 2 and Figures 5 to 12 As shown), then yw1=0, as Figure 6 As shown.

[0113] Furthermore, the coordinates zw1 of the first virtual reference point w1 along the z-axis can be determined from the coordinates of the first position along the z-axis, thereby determining the lateral distance in the image reference frame. In other words, the virtual reference point w1 can have the same height along the z-axis as the detected object OBJ1. In one embodiment, the height corresponding to the height zw1 of the virtual reference point w1 along the z-axis can be predefined and stored by a device 2 corresponding to a predefined ground height. In another embodiment, the height zw1 can also be different from the height of the detected object OBJ1.

[0114] Based on such a virtual reference point w1 located on the acquired image and the current size associated with the vehicle and its neighboring vehicle VV1. (Here, the dimensions L1, L2, and H are customized from the previously defined ones), a virtual object OV1 with coordinates (yw1, zw1) in a two-dimensional space (y, z) can be created. Specifically, the virtual object OV1 can substantially include the first detected object OBJ1, since the latter is identified as corresponding at least partially to the adjacent vehicle VV1 detected in the acquired image. For example, Figure 6 The virtual object OV1 is represented in the image. The created virtual object OV1 then corresponds to modeling the occupied space in the first acquired image of the neighboring vehicle VV1 detected based on the first detected object OBJ1 on the acquired image (and therefore in the image reference frame (y,z)). The current position of the neighboring vehicle VV1... It can be used to refer to the location of reference point W1 or virtual reference location w1.

[0115] Optionally, the coordinates (yw1, zw1) of the virtual object OV1 created thus can be compared with the parameters ( , , The detected neighboring vehicle VV1 and the first acquisition time T1 are fed into a Kalman filter in association. Therefore, prediction parameters associated with the created virtual object OV1 can be estimated as the object of interest (typically a neighboring vehicle VV1) corresponding to the virtual object OV1 moves through the environment ENV. Specifically, the lateral distance of the object... It can be associated with a virtual object OV1. For example, at the first acquisition time T1 that enables the creation of a virtual object OV1 from the first detected object OBJ1, the lateral distance of the object. Can be with the first lateral distance Correspondingly. Then, the horizontal distance of the object. The value can change depending on the continuous predictions of a data estimator that estimates the motion of the neighboring vehicle VV1 being tracked (and modeled by the virtual object OV1).

[0116] Optionally, a virtual object OV1 associated with the detected neighboring vehicle VV1 can be displayed on the communication interface 50 corresponding to the display screen, for example, by overlaying the virtual object OV1 onto the display of the image stream from camera 1. Figure 6 As shown. Therefore, the driver of the primary vehicle VP can quickly identify the space occupied by the adjacent vehicle VV1 detected in their field of view (FOV). Specifically, this is similar to the initial position associated with the detected object OBJ1 and the lateral distance associated with the detected object OBJ1. Or parameters associated with adjacent vehicle VV1 ( , , ),like Figure 6 As shown, such a virtual object OV1 is associated with a given acquisition time—here, the first acquisition time T1—which is linked to a certain location and certain kinematic conditions of the adjacent vehicle VV1 in the environment ENV, which are estimated based on the object OBJ1 detected at the first acquisition time T1.

[0117] Now we consider Figure 4 The method for early tracking at a second acquisition time T2, which is after the first acquisition time T1, is illustrated. For example, after the initial implementation of the above steps has created a virtual object OV1 associated with the first acquisition time T1 and a first detected object OBJ1, the second acquisition time T2 is considered. At the second acquisition time T2, the second object OBJ2 can be detected in step 410. For example, such a second object OBJ2 can be detected in a second image acquired at the second acquisition time T2 in step 400, which is different from the first image in which the first object OBJ1 was detected. For example, Figure 8 The second object OBJ2 is shown being detected in the second image. For comparison, the first object OBJ1 associated with the first acquisition time T1 is also shown in the figure.

[0118] As described above, a second lateral distance associated with the second detected object OBJ2 can be determined in step 420. And in step 430, a second reference point W2 and parameters associated with the second detected object OBJ2 can be determined. , ,).

[0119] In step 440, object verification is performed on the second detected object OBJ2 to determine whether the second detected object OBJ2 can be associated with an existing (already created) virtual object. Specifically, in step 440, it can be determined that the second detected object OBJ2 at the second acquisition time T2 can be associated with the virtual object based on the parameters determined in step 430. , This is associated with the movement of the adjacent vehicle VV1 previously detected by device 2 (i.e., at the first acquisition time T1), and is represented by a virtual object OV1, such as... Figure 6 , Figure 7 and Figure 9 As shown.

[0120] At step 450, the parameters associated with the second detected object OBJ2 ( , ) can also be fed into the Kalman filter. Therefore, the parameters predicted by the Kalman filter for the virtual object OV1 ( , , Specifically, parameters associated with the second detected object OBJ2 can be considered. , ).

[0121] The step 470 of updating the virtual object OV1 associated with the adjacent vehicle VV1 at the second acquisition time T2 will now be described in detail. This step 470 of updating the virtual object OV1 specifically allows the parameter (p) to be obtained at the end of step 470. n+1 * ,v n+1 * ,s n+1 * The set of parameters allows characterizing the position and motion of the virtual object OV1 (and therefore the adjacent vehicle VV1) in the environment ENV at the second acquisition time T2. The virtual object OV1 to be updated is compared with the current parameters to be updated ( , , (related to)

[0122] Step 470 of updating virtual object OV1 specifically considers the state and predicted (or future) parameters (p) of virtual object OV1 provided by the data estimator. n+1 , v n+1 , s n+1 ). Such prediction parameters (p) n+1 , v n+1 , s n+1This reflects, for example, an estimate of the state of the adjacent vehicle VV1 provided by the Kalman filter of device 2, which is predicted based on data acquired during the first acquisition time T1. Specifically, the predicted state of the virtual object OV1 may also include the predicted lateral distance d of the object. lat,OV1 The predicted lateral distance corresponds to the lateral spacing between the neighboring vehicle VV1 and the master vehicle VP estimated by the Kalman filter at the second acquisition time T2.

[0123] In the context of the proposed method, step 470 of updating virtual object 407 includes sub-steps 471 and 472 of updating the length of virtual object OV1 according to a first predefined criterion and sub-steps 473 and 474 of updating the position of virtual object OV1 according to a second predefined criterion.

[0124] In substep 471, the predicted or future parameters (p) of the virtual object OV1 estimated by the Kalman filter are... n+1 , v n+1 , s n+1 ) and the actual parameters of virtual object OV1 ( , , The comparison is performed. Specifically, the current position p associated with the virtual object OV1 is compared. n Compared with the estimated future position p for virtual object OV1 n+1 Compare. If in substep 471, at the current position p n With future position p n+1 If such a comparison satisfies the first criterion, then the current size s of the virtual object OV1 is realized. n And specifically the current length s n (X) Update 472. Specifically, if the predicted future position p of virtual object OV1 n+1 Indicates that the virtual object OV1 has been relative to its current position p n Moving backward within the environment ENV satisfies the first criterion. For example, if the predicted position p of virtual object OV1... n+1 The lateral component of (X) is smaller than its current position p in the reference frame (X,Y,Z). n If the horizontal component of (X) is given, then the virtual object can be considered to have information about its current position p. n Further back in the sequence. In this case, device 2 maintains the state of the virtual object OV1 by systematically extending the length of the virtual object OV1 associated with the adjacent vehicle VV1, but its current size s n except.

[0125] In substep 472, the current length s of the virtual object OV1 is then implemented. n The update then updates the current size of the virtual object OV1 associated with the adjacent vehicle VV1 by modifying the length associated with it. Such lengths are, for example, based on dimensions. X coordinate: in: in: - It is a difference in length (expressed in meters). - It is the updated length of the adjacent vehicle VV1 in the world reference frame. - It is the current length of the adjacent vehicle VV1 in the world reference frame. - The coordinates of the reference point W2 in the world reference frame along the X-axis are predicted by the Kalman filter. - It is the coordinate of the position of the first reference point W1 along the X-axis in the world reference frame.

[0126] Specifically, length This corresponds to the longitudinal component (along the X-axis) of the size associated with the virtual object OV1.

[0127] In other words, sub-step 472 involves extending the length of the virtual object OV1 modeled for the adjacent vehicle VV1. Then, the reference point W associated with the front side of the adjacent vehicle VV1 is positioned in the actual location. With the predicted location The positional difference between them extends to the length middle.

[0128] Specifically, such an extension in step 472 is achieved by the following: in: - It is the updated reference point associated with the updated virtual object OV*. - It is the first reference point associated with the virtual object OV1 before the update.

[0129] In other words, this achieves the update of the length of adjacent vehicle VV1 in step 472 while maintaining the position of the front side of the virtual object OV*. Therefore, updating the virtual object OV1 to OV* allows the length of adjacent vehicle VV1 to be updated by preventing the virtual object OV1 from "moving backward" in the acquired image.

[0130] Specifically, in substep 472, the size s of the virtual object is updated. n Make the current speed Keep it constant.

[0131] Therefore, the updated parameters at the end of substep 472 ( , )satisfy: Figure 7 This shows the length of the virtual object OV1 associated with the adjacent vehicle VV1 in substep 472. Update.

[0132] If the first criterion is not met in substep 471, the length update is not performed, and step 470—updating virtual object OV1—continues with substeps 473 and 474.

[0133] In substep 473, if the second criterion is met, the position p of the virtual object OV1 is determined. n The update. This second criterion specifically depends on the second lateral distance d determined from the second detected object OBJ2. lat,2 The lateral distance d between the object and the virtual object OV1 is estimated. lat,OV1 The second comparison between them. Specifically, if in substep 473, the second lateral distance d lat,2 Strictly less than the horizontal distance d of the object lat,OV1 Then the current position p of the virtual object OV1 is realized. n In update sub-step 474, virtual object OV1 is repositioned such that its updated object lateral distance d lat,OV1 *Minimum (i.e., d) lat,OV1 *= min(d lat,OV1 , d lat,2 In fact, this update 474 corrects the lateral distance d between the adjacent vehicle VV1 and the master vehicle VP. lat The lateral distance of virtual object OV1 was overestimated in previous iterations, and its position needs to be updated. Then, the current position of the adjacent vehicle VV1 is updated. The update specifically includes updating the current position of at least the adjacent vehicle VV1. lateral component (i.e., along the Y-axis in the world reference frame (X,Y,Z). In one embodiment, for the current position p n The update can also include updates to the current position p n Other components (e.g., height component p) n (Z)) update.

[0134] In substep 474, if the second criterion is satisfied, updating such a current position involves updating the reference point. Associated with the updated virtual object OV*, where such an updated reference point With the first reference point Different. Then, based on the second lateral distance... To determine the updated reference point Such an updated reference point The positioning specifically follows the guidelines regarding the first reference point. The constraints on the image are the same as the constraints.

[0135] Specifically, in substep 474, the parameters are updated to maintain a constant image flow.

[0136] Therefore, the updated parameters at the end of substep 474 ( , It can satisfy: Specifically, the updated parameters at the end of sub-step 474 ( , The representation of ) depends on the position. And the definitions of reference points w1 and w2*. For example, if point Corresponding to the position of the first reference point w1 located in the middle of the virtual object OV1, the updated parameters at the end of substep 474 ( , It can satisfy the following relationship: In one embodiment, the size can also be updated in sub-step 474. For example, to maintain a constant footprint for the updated virtual object OV* in the image: Figure 9 This illustrates an update 474 for obtaining the updated virtual object OV* based on the position of virtual object OV1, where the second criterion (second lateral distance d) is satisfied. lat,2 Less than the horizontal distance d of the object lat,OV1 ).

[0137] Therefore, update 474 modifies the position p of the virtual object OV1. n Horizontal distance d of the object lat,OV1 And therefore the velocity v n At the same time, it sets a parameter of the virtual object OV1 (such as collision time, ITTC, or...). () Remain constant.

[0138] If in substep 473, the horizontal distance d lat,OV1 Less than the second lateral distance d lat,2 Then, the position update 474 of the virtual object OV1 and the direct prediction of the virtual object OV1 based on the Kalman filter estimation can be implemented in the regular update step 475 of the virtual object OV1. n+1 , v n+1 , s n+1 Update of ).

[0139] therefore, Figure 4 The method for early tracking described herein allows for early tracking of adjacent vehicles VV1 before their front side becomes visible in the field of view (FOV). Specifically, the proposed method allows for specific updates to the length and / or position of a virtual object OV1 (which models adjacent vehicle VV1) to compensate for and reduce overestimations of the lateral distance between adjacent vehicle VV1 and the master vehicle VP, which can lead to hazardous situations (e.g., an overestimation of the lateral distance makes adjacent vehicle VV1 appear farther from the master vehicle than it actually is). Furthermore, the length of the virtual object OV1 can be updated even if no new object is detected (e.g., a second detected object OBJ2).

[0140] Optionally, the proposed method for early tracking can be combined with tracking implemented by a classifier. In other words, even when the front side of the adjacent vehicle VV1 becomes visible (e.g., in...), the tracking can continue even when the front side of the adjacent vehicle VV1 becomes visible. Figure 10 It can also be achieved at the third acquisition time T3 in the data collection process. Figure 4The method for early tracking described herein. Alternatively, the method for early tracking can be implemented as long as the neighboring vehicle VV1 cannot be tracked with sufficient reliability—for example, as long as the front end of the neighboring vehicle VV1 is not yet visible (e.g., in the scenarios of the first acquisition phase T1 and the second acquisition phase T2). To this end, the parameters fed into the Kalman filter in step 440 can be correlated with a selected variance value, such that when the parameters estimated by the front view classifier are fed into the Kalman filter, the parameters estimated using the proposed method for early tracking are no longer important to the filter's predicted parameters. Therefore, the method can be repeated as long as an image is acquired by device 2 or as long as the parameters fed into the Kalman filter are used to predict parameters (e.g., as long as data from the classifier with low uncertainty and therefore low variance or covariance has not yet been fed into the Kalman filter). Figure 4 The method shown.

Claims

1. A method for tracking at least one neighboring vehicle (VV1) in an environment (ENV) of a master vehicle (VP), wherein the neighboring vehicle (VV1) and the master vehicle (VP) are motor vehicles, the method being implemented by a device (2) configured to provide driver assistance functions for the master vehicle (VP), the device (2) being connected to at least one camera (1) on the master vehicle (VP) and capable of acquiring images of the scene surrounding the master vehicle (VP) according to at least one field of view (FOV) at acquisition times (T1, T2, T3). The method includes the following steps: - Detect (410) a first object (OBJ1) having a vertical plane extending laterally relative to the master vehicle (VP) based at least on a first image acquired at the first acquisition time (T1). - Determine at least one data item related to the first position associated with the detected first object (OBJ1) based on a predefined coordinate system (X,Y,Z). - Create (460) a virtual object (OV1) associated with a set of parameters (pn, vn, sn), the parameters including at least the current position (pn) associated with the virtual object (OV1), the current position (pn) being estimated (430) based at least on the data item associated with the first position according to the predefined coordinate system. - Estimate (440) the future position (pn+1) of the virtual object (OV1) based at least on the current position (pn) of the virtual object, and - Update the parameters associated with the virtual object (OV1) at least based on the future position (pn+1), wherein the update (470, (pn+1*, vn+1*, sn+1*) of the parameters associated with the virtual object (OV1) includes updating the current length (sn(X)) of the virtual object (OV1) if the first criterion (471) is satisfied.

2. The method of claim 1, wherein updating the current length of the virtual object (OV1) comprises extending the length of the virtual object (OV1) by a value determined from the interval (δ) between the current position (pn) and the future position (pn+1), while keeping the current position (pn) of the virtual object (OV1) constant.

3. The method according to any one of the preceding claims, wherein the update of the current length of the virtual object (OV1) is performed if a first comparison between the future position (pn+1) and the current position (pn) indicates a backward position of the virtual object (OV1) relative to the current position (pn) in the acquired first image.

4. The method according to any one of the preceding claims, wherein the date associated with the first location includes a first lateral distance (dlat,1) associated with the first detected object (OBJ1) for the first acquisition time (T1), the first lateral distance (dlat,1) being determined from measurements of the first detected object (OBJ1) in the first acquired image.

5. The method according to any one of the preceding claims, wherein the future position (pn+1) is associated with the lateral distance (dlat, OV1) of the object from the virtual object (OV1).

6. The method according to the preceding claim, further comprising: - Obtain a data item associated with a second location of the second detected object (OBJ2) from at least a second image acquired at a second acquisition time (T2) after the first acquisition time (T1), the data item associated with the second location including a second lateral distance (dlat,2) associated with the second detected object (OBJ2) for the second acquisition time (T2). - If the second detected object (OBJ2) is determined to belong to the virtual object (OV1), then in the second comparison, the second lateral distance (dlat,2) and the lateral distance (dlat,OV1) of the object are compared. The update (470, (pn+1*,vn+1*,sn+1*)) of the parameters associated with the virtual object (OV1) further depends on the result of the second comparison.

7. The method according to claim 6, wherein, If the second lateral distance (dlat,2) is strictly less than the lateral distance (dlat,OV1) of the object, then the update (470, (pn+1*,vn+1*,sn+1*) of the parameters associated with the virtual object (OV1) includes updating (474, pn+1*) the current position (pn) of the virtual object (OV1), and the update (474, pn+1*) of the current position (pn) includes at least updating the lateral component (pn(Y)) of the current position (pn) of the virtual object (OV1), the lateral component corresponding to the repositioning of the virtual object (OV1).

8. The method according to any one of the preceding claims, wherein the current position (pn) of the virtual object (OV1) is associated with the position of a first reference point (w1) belonging to the virtual object, and the position of the first reference point (w1) belonging to the virtual object (OV1) is determined such that, in the predefined coordinate system corresponding to the image reference system (x,y) in the acquired image, the first reference point (w1) is located on the vertical edge of the first image.

9. The method according to any one of the preceding claims, wherein the parameters (pn, vn, sn) associated with the virtual object (OV1) are updated (470, (pn+1*, vn+1*, sn+1*) while maintaining constant parameters of the virtual object (OV1). ).

10. An apparatus (2) configured to provide driver assistance functions for a primary vehicle (VP), the apparatus (2) being connected to at least one camera (1) on the primary vehicle (VP) and capable of acquiring images of the scene surrounding the primary vehicle (VP) according to at least one field of view (FOV) at acquisition times (T1, T2, T3), wherein the apparatus (2) includes at least one processing circuit (20, 30, 40, 50, 60) configured to implement a method for tracking at least one adjacent vehicle (VV1) present in the environment (ENV) of the primary vehicle (VP) according to any one of claims 1 to 9.

11. A computer program comprising instructions for implementing the method according to any one of claims 1 to 9 when the program is executed by at least one processor (21, 31, 41, 51, 61).

Images