Situation awareness processor

[0002] The instant application claims the benefit of U.S. Provisional Application Serial No. 60/210,878 filed on Jun. 9, 2000 (5701-00266).

[0003] The instant application is related to U.S. application Ser. No. 09/877,493, entitled Track Map Generator, filed on Jun. 8, 2001 (5701-01265),

[0004] The above-identified applications are incorporated herein by reference.

[0005] In the accompanying drawings:

[0006] FIG. 1 illustrates a target in view of a host vehicle in a local coordinate system of the host vehicle;

[0007] FIG. 2 illustrates a block diagram system incorporating a situation awareness processor;

[0008] FIG. 3 illustrates a block diagram of host vehicle with a system incorporating a situation awareness processor;

[0009] FIG. 4 illustrates a general inference process of a situation awareness processor; and

[0010] FIG. 5 illustrates an example of an inference process of a situation awareness processor.

[0011] There exists a need for an improved predictive collision sensing or collision avoidance system for automotive applications that can sense and identify an environment of a host vehicle with sufficient range and accuracy so that proper countermeasures can be selected and taken sufficiently early to either avoid a collision; or to mitigate injury therefrom either to occupants of the host vehicle, or to pedestrians outside thereof. As used herein, the term predictive collision sensing system will also refer to a collision avoidance system, so as to mean a system that can sense and track targets in the environment of the host vehicle, and then either suggest, or automatically invoke, countermeasures, that would improve safety. Generally, a predictive collision sensing system tracks the motion of the host vehicle relative to its environment, or vice versa, for example, using a radar system with an associated target tracker. The environment may include both stationary and moving targets. An automotive environment is distinguished from other target tracking environments—for example, that of air or sea vessels—in that automotive vehicles are primarily operated in an environment that is constrained by roadways. There are, of course, exceptions to this, for example, parking lots or off-road driving conditions, but these exceptions generally account for a relatively small percentage of vehicular operating time, or for a relatively small risk of collisions for which there would be benefit from a predictive collision sensing system.

[0012] Referring to FIGS. 1-3, a situation awareness processor 10 is illustrated in a radar processing system 12 that processes radar data from a radar system 14 incorporated in a host vehicle 16. Generally, the host vehicle 16 incorporates a sensor for sensing targets 18 in the environment thereof, and incorporates a system for actuating and/or controlling associated countermeasures responsive to the relative motion of the host vehicle 16 with respect to one or more targets. The sensor for sensing targets 18 is, for example, a radar or lidar sensor system that senses and tracks the location of targets 18 relative to the host vehicle 16, and predicts if a collision between the host vehicle 16 and a target 18 is likely to occur, for example, as disclosed in commonly owned U.S. Pat. No. 6,085,151 that is incorporated herein by reference.

[0013] Referring to FIG. 1, the host vehicle 16 is seen traveling in the X-direction at a host speed 20 in the environment of one or more targets 18, each at an associated bearing 22 and distance 24 relative to the host vehicle 16, each having an associated target speed 26 at an associated heading 28. Generally, the targets 18 may be either stationary, or moving, relative to the environment 30 of the host vehicle 16.

[0014] FIG. 2 illustrates a block diagram of the radar processing system 12, wherein each block in the block diagram comprises associated software modules that receive, prepare and/or process data provided by a radar system 14 mounted in the host vehicle. Data from the radar system 14 is preprocessed by a preprocessor 32 so as to generate radar data, for example, range, range rate, and azimuth angle, suitable for target tracking. The radar data typically covers a relatively wide field of view forward of the host vehicle, at least +/−5 degrees from the host vehicle's longitudinal axis, and possibly extending to +/−180 degrees, or larger, depending upon the radar and associated antenna configuration. A present exemplary system has a field of view of +/−55 degrees.

[0015] A tracker 34 converts radar output data (range, range rate & azimuth angle) into target speed and X-Y coordinates specifying the location of a target. For example, the system of U.S. Pat. No. 6,085,151 discloses a system for tracking multiple targets, and for clustering associated radar data for a single target. An associator 36 relates older track data to that from the latest scan, compiling a track history of each target.

[0016] A track map generator 38 generates a track map comprising a grid of heading and quality information from the track data as a record of the evolution of target motion relative to the host vehicle. The track map is updated with track data from subsequent targets 18 that enter the field-of-view of the host vehicle 16, and old data fades from the map with time. Accordingly, the track map provides a representation of the paths followed by targets 18 relative to the host vehicle 16, which paths are normally constrained by the associated roadways in the environment 30 of the host vehicle 16. For example, a track map generator 38 is disclosed in a separately filed U.S. application Ser. No. 09/877,493, entitled Track Map Generator, filed on Jun. 8, 2001.

[0017] A situation awareness processor 10 uses 1) the track map, 2) data acquired indicative of the position and/or motion of the host vehicle 16, i.e. host vehicle information 40, and possibly 3) environment data 42, to determine the most likely or appropriate driving situation from a set of possible driving situations. For example, the environment data 42 can include navigation data from a navigation system 44, digital maps, real-time wireless inputs of highway geometry and nearby vehicles, and data from real-time transponders such as electromagnetic or optical markers built into highways; and can be used—along with target tracking data from the tracker 34/associator 36—by a road curvature estimator 46 to provide an estimate of road curvature to the situation awareness processor 10. For example, navigation data such as the location and direction of a vehicle can be measured by a GPS receiver; by a dead reckoning system using measurements of vehicle heading from a compass or directional gyroscope, and vehicle distance and heading from wheel speed or rotation measurements, in conjunction with a map matching algorithm; or a combination thereof.

[0018] The situation awareness processor 10 stores and interprets the track map from the track map generator 38, and compares the progress over time of several target tracks. Evaluation of the relative positions and progress of the tracked targets permits identification of various driving situations, for example a location situation, a traffic situation, a driving maneuver situation, or the occurrence of sudden events. Several approaches can be used to identify the situation, for example set theoretic reasoning (using for example random set theory or evidential reasoning); Bayesian inference; or a neural network.

[0019] Examples of location situations include a divided or undivided highway, an intersection, a freeway entrance or exit, a parking lot or off-highway situation, and a stopped object on the left or right of the host vehicle. Examples of traffic situations include crowded traffic, loose traffic, or normal traffic. Examples of driving maneuver situations include target cut-in, host vehicle lane or speed changing, or target speed changing.

[0020] The situation estimated by the situation awareness processor 10, together with collision and crash severity estimates from a collision estimator 48 and a crash severity estimator 50 respectively, are used as inputs to a response generator 52 to select an appropriate countermeasure 54, for example, using a decision matrix. The decision of a particular response by the response generator 52 may be based on, for example, a rule-based system (an expert system), a neural network, or another decision means.

[0021] Examples of countermeasures 54 that can be activated include, a warning device to warn the driver to take corrective action, for example 3D audio warning (for example, as disclosed in commonly owned U.S. Pat. No. 5,979,586 that is incorporated by reference herein); various means for taking evasive action to avoid a collision, for example the engine throttle, the vehicle transmission, the vehicle braking system, or the vehicle steering system; and various means for mitigating injury to an occupant if a collision is unavoidable, for example a motorized safety belt pretensioner, or internal or external airbags. The particular one or more countermeasures 54 selected, and the manner by which that one or more countermeasures 54 are activated, actuated, or controlled, depends up the situation identified by the situation awareness processor 10, and upon the collision and crash severity estimates. By way of example, one potential scenario is that the response to encroachment into the host's lane of travel would be different depending upon whether the target is coming from the opposite direction or going the same way as the host vehicle 16, but cutting into the lane thereof. By considering the traffic situation giving rise to the threat, the countermeasures 54 can be better adapted to mitigate that threat. By using a radar system, or generally a predictive collision sensing system, to sense targets within range of the host vehicle, the countermeasures 54 may be implemented prior to an actual collision so as to either avoid a collision, or to mitigate occupant injury therefrom.

[0022] Referring to FIG. 3, the radar system 14 and radar processing system 12 are seen incorporated in host vehicle 16, wherein the host vehicle information 40 is provided by associated sensors, for example, operatively connected to a vehicle processor 56 that provides the resulting host vehicle information 40 to the situation awareness processor 10. For example, one or more wheel speed sensors 58 can provide wheel speed measurements from one or more associated wheels of the vehicle, the average of which, for laterally opposed wheels, provides a measure of vehicle speed; the difference of which provides a measure of yaw rate. Alternately, or additionally, the vehicle speed may be measured from the rotation of a transmission output shaft. A steering angle sensor 60 operatively connected to the steering shaft provides a measure of steering angle responsive to the turning of the steering wheel by the driver. The yaw rate of the host vehicle 16 may alternately, or additionally, be sensed by a gyro 62, wherein the integration of yaw rate provides a measure of the heading of the host vehicle 16.

[0023] The situational awareness processor 10 transforms the sensor information collected by various sensors into a coherent understanding of the driving environment, such that a more proper response can be invoked responsive to an emergency. Stated in another way, the situational awareness processor 10 interprets the meaning of sensor data. The inputs to the situational awareness processor 10 are transformed from a “state” domain into an “event” domain. Whereas variables in state domain are usually represented numerically, the variables in event domain are commonly represented as events and by a confidence of the occurrence thereof. The situational awareness processor 10 identifies an “event” corresponding to particular a “situation”. Tables 1-3 provide examples, and associated definitions, of various events.

[0024] Inputs for identifying relevant events are obtained by processing the original sensor reports including radar reports, host vehicle speed and yaw rate, GPS host location report and digital map report. These inputs include, but are not limited to, those listed in Table 4, the information of which, except for host absolute location, type of road and approaching road structure, is interrelated by the coordinate system of FIG. 1. Host absolute location is defined in the coordinate system of the digital map, and the type of road and approach road structures are expressed in terms of events.

[0025] The output of the situational awareness processor 10 comprises a list of identified events, and associated confidence levels. The events collectively define the current driving situation. The confidence can be represented in several ways, such as probability and possibility, each generated by a different inference (reasoning) mechanism, as discussed further hereinbelow. The outputs are determined by an inference process, which is a recursive decision making process that correlates events with their associated and corresponding inputs. A decision is given in terms of a confidence for the occurrence of a particular event. A general formula for calculating the confidence of an event can is given as:

[0026] p_k=f(p_k-1,e₁,e₂,K,e_n)  (1)

[0027] where p_kis the confidence level, and e_iare the inputs, at time k.

[0028] Various methods can be used in the inference engines that specify the recursive updating mechanism. Among them, the most widely used are 1) Dempster-Shafer Evidential Reasoning, 2) Fuzzy Logic Reasoning, 3) Bayesian Inference, and 4) Neural Network. Random set theory can also used in place of the evidential reasoning or fuzzy logic reasoning as a more general approach.

[0029] In preparation for using a particular inference engine, the above-described events are first grouped into several classes, each containing a number of mutually exclusive atomic events. Inference is made within each class, although the output of one inference engine can be fed back as an input to another inference engine. A decision on the nature of the current driving situation is made responsive to the collective output of all inference engines. For example, Table 5 lists a set of reorganized event classes based upon the events listed in Tables 1-3.

[0030] The decision process based on the reorganized events essentially forms a networked inference engine, which is illustrated generally in FIG. 4, and which is illustrated more specifically in FIG. 5.

[0031] By way of example, the decision process is now illustrated by deriving a Bayesian inference of Approaching Road Structure. The Bayesian formula is given by: $\begin{array}{ccc}p\left(A_i|B\right)& =& \frac{p\left(B|A_i\right)p\left(A_i\right)}{p\left(B\right)}\\ & =& \frac{p\left(B|A_i\right)p\left(A_i\right)}{\sum _{j=1}^{n}p\left(B|A_j\right)p\left(A_j\right)}\end{array}$

[0032] Denoting e_κ, κ=a, b, Λ k, as all the information available to the inference engine, and {e_κ} as the collection of all available inputs, e_acollectively refers to all the information carried by the track map, which can be further decomposed into a set of G grids, i.e. e_a{e_ah^ξ,e_aq^ξ,ξ=1, 2, Λ G}, where e_ah^ξ and e_aq^ξ are heading and quality information of grid ξ, respectively. For N targets, then e_e=N. Superscripts can be used to distinguish targets, e.g. e_b^r, e_c^rand e_d^rrepresent speed, heading, and location of target r, r=1, 2 . . . N, respectively. Of the target information, each e_d^rcontains two elements, defining a target's location in X and Y coordinates, i.e. e_d^r=(x^r, y^r) Road type and approaching road structure are events in nature. Each event is from a class of possible events as defined earlier. Generally, a superscript can be used to differentiate an event from peers in its class, e.g. e_i^α and e_j^β specifies a road type and an approaching road structure, respectively.

[0033] The output of the inference network comprises the seven events from the seven classes defined earlier, wherein event is denoted as S, subscripts are used to denote the class of an event, and superscript are used to denote the intended event of the class specified by the subscript. Accordingly, S_η^λ is an output event for class η and event λ in the class. For example, if η=1, λ=2, then S_η^λ means freeway entrance.

[0034] Given the above definitions, the Bayesian inference for Approaching Road Structure can be derived as follows. Under Bayesian framework, the outcome of an inference is the posterior probability of a given event. For approaching road structures, the probabilities to be evaluated are S₁^λ, λ=1, 2, 3, 4, given the collection of evidence {e_κ} and knowledge of a related event in class S₂, are given by: $p\left(S_1^{\lambda }|S_2^{{\lambda }_0},\left\{e_{\kappa }\right\}\right)=\frac{p\left(\left\{e_{\kappa }\right\}|S_1^{\lambda },S_2^{{\lambda }_0}\right)p\left(S_1^{\lambda }|S_2^{{\lambda }_0}\right)}{\sum _{{\lambda }_1=1}^{4}p\left(\left\{e_{\kappa }\right\}|S_1^{{\lambda }_1},S_2^{{\lambda }_0}\right)p\left(S_1^{{\lambda }_1}|S_2^{{\lambda }_0}\right)}$

[0035] wherein, λ₀=1, 2, Λ 6 is the index for events belonging to the road type class.

[0036] Using the previously obtained p(S₁^λ|S₂^λ, {e_κ}) to approximate current p(S₁^λ|S₂^λ), provides the following recursive probability updating equation: $p_k\left(S_1^{\lambda }|S_2^{{\lambda }_0},\left\{e_{\kappa }\right\}\right)=\frac{p\left(\left\{e_{\kappa }\right\}|S_1^{\lambda },S_2^{{\lambda }_0}\right)p_{k-1}\left(S_1^{\lambda }|S_2^{{\lambda }_0},\left\{e_k\right\}\right)}{\sum _{{\lambda }_1=1}^{4}p\left(\left\{e_{\kappa }\right\}|S_1^{{\lambda }_1},S_2^{{\lambda }_0}\right)p_{k-1}\left(S_1^{{\lambda }_1}|S_2^{{\lambda }_0},\left\{e_k\right\}\right)}$

[0037] wherein p({e_κ}|S₁^λ, S₂^λ) is the probability density of set {e_κ} given the condition that events S₁^λ and S₂^λare true. Considering the individual elements in {e_κ} gives: $p\left(\left\{e_{\kappa }\right\}|S_1^{\lambda },S_2^{{\lambda }_0}\right)=p\left(e_a,\left\{e_b^r,e_c^r,e_d^r\right\},e_f,e_i^{\alpha },e_j^{\beta }|S_1^{\lambda },S_2^{{\lambda }_0}\right)=p\left(e_a|S_1^{\lambda },S_2^{{\lambda }_0}\right)p\left(e_f|S_1^{\lambda },S_2^{{\lambda }_0}\right)p\left(e_i^{\alpha }|S_1^{\lambda },S_2^{{\lambda }_0}\right)p\left(e_j^{\beta }|S_1^{\lambda },S_2^{{\lambda }_0}\right)\times \prod _{r=1}^{N}p\left(e_b^r|S_1^{\lambda },S_2^{{\lambda }_0}\right)p\left(e_c^r|S_1^{\lambda },S_2^{{\lambda }_0}\right)p\left(e_d^r|S_1^{\lambda },S_2^{{\lambda }_0}\right)$

[0038] wherein individual conditional probabilities/probability densities can be found as: $\begin{array}{cc}p\left(e_a|S_1^{\lambda },S_2^{{\lambda }_0}\right)p\left(e_a|S_1^{\lambda },S_2^{{\lambda }_0}\right)=\prod _{\xi =1}^G\left[p\left(e_{\mathrm{ah}}^{\xi }|S_1^{\lambda },S_2^{{\lambda }_0},\xi =\mathrm{TRUE}\right)p\left(\xi =\mathrm{TRUE}|S_1^{\lambda },S_2^{{\lambda }_0}\right)+p\left(e_{\mathrm{ah}}^{\xi }|S_1^{\lambda },S_2^{{\lambda }_0},\xi =\mathrm{FALSE}\right)p\left(\xi =\mathrm{FALSE}|S_1^{\lambda },S_2^{{\lambda }_0}\right)\right]\mathrm{As}p\left(\xi =\mathrm{TRUE}|S_1^{\lambda },S_2^{{\lambda }_0}\right)\text{]}=e_{\mathrm{aq}}^{\xi }\mathrm{and}p\left(\xi =\mathrm{FALSE}|S_1^{\lambda },S_2^{{\lambda }_0}\right)\text{]}=1-e_{\mathrm{aq}}^{\xi },\mathrm{then}& \left(1\right)\end{array}$ $p\left(e_a|S_1^{\lambda },S_2^{{\lambda }_0}\right)=\prod _{\xi =1}^G\left[p\left(e_{\mathrm{ah}}^{\xi }|S_1^{\lambda },S_2^{{\lambda }_0},\xi =\mathrm{TRUE}\right)e_{\mathrm{aq}}+p\left(e_{\mathrm{ah}}^{\xi }|S_1^{\lambda },S_2^{{\lambda }_0},\xi =\mathrm{FALSE}\right)\left(1-e_{\mathrm{aq}}^{\xi }\right)\right].$

[0039] A suitable way to model p(e_ah^ξ|S₁^λ, S₂^λ, ξ=TRUE) is to a use Guassian mixture, i.e. an approximation involving a summation of several Guassian distributions: $p\left(e_{\mathrm{ah}}^{\xi }|S_1^{\lambda },S_2^{{\lambda }_0},\xi =\mathrm{TRUE}\right)=\frac{1}{N_f}\sum _{i=1}^{N_f}\frac{1}{2{\mathrm{\pi \sigma }}_i}{e}^{\frac{{\left(e_{\mathrm{ah}}^{\xi }-{\mu }_i\right)}^2}{2{\sigma }_i^2}}$

[0040] where N_𝑓 is the number of features selected for an approaching road structure. For example, the following parameters may be chosen for an intersection: N_𝑓=3, μ₁=0°, μ₂=90°, μ₃=90°, and σ₁=σ₂=σ₃=10°.

[0041] The distribution p(e_ah^ξ|S₁^λ, S₂^λ, ξ=FALSE) may be modeled as a uniform distribution, e.g., $p\left(e_{\mathrm{ah}}^{\xi }|S_1^{\lambda },S_2^{{\lambda }_0},\xi =\mathrm{FALSE}\right)=\frac{1}{180°}.$

[0042] (2) p(e_b^r|S₁^λ, S₂^λ), p(e_c^r|S₁^λ, S₂^λ), p(e_d^r|S₁^λ, S₂^λ) and p(e_𝑓|S₁^λ, S₂^λ) can be handled similar to p(e_a|S₁^λ, S₂^λ).

[0043] (3) p(e_i^α|S₁^λ, S₂^λ):

[0044] Probability of event-type input e_i^α can be decided as $p\left(e_i^{\alpha }|S_1^{\lambda },S_2^{{\lambda }_0}\right)=\{\begin{array}{cc}p\left(e_i^{\alpha }=\mathrm{TRUE}\right),& \mathrm{if}{S}_2^{{\lambda }_0}=e_i^{\alpha }\\ 1-p\left(e_i^{\alpha }=\mathrm{TRUE}\right),& \mathrm{if}{S}_2^{{\lambda }_0}\ne e_i^{\alpha }\end{array}$

[0045] Where p(e_i^α=TRUE) can be obtained by the quality measure of GPS/map matching.

[0046] (4) p(e_i^β|S₁^λ, S₂^λ) can be handled similar to p(e_i^α|S₁^λ, S₂^λ).

[0047] The actual forms and parameters of above probability terms would be obtained based on experiments, with simplification to reduce computational burden.

[0048] The above inference engine can be replaced by any of the other three widely used reasoning techniques, namely fuzzy logic reasoning, Dempster-Shafer evidential reasoning and neural networks. Particularly, a collection of atomic events can be specified based on the classes of events to be recognized, for example, as are illustrated in Table 6.

[0049] Note that the defined atomic events are not exactly the same as the situations. Some are intersections of several events, some are unions of events, some are actually part of a situation, and some have not been defined. However, these atomic sets collectively can represent the situations defined earlier.

[0050] Given the defined sets of atomic events, the output events can be obtained by joining the atomic events, as illustrated in Table 7.

[0051] The Dempster-Schafer mass functions assigned to the events can be obtained based on experiments. Mass updating can also obtained based on evidential reasoning theory.

[0052] While specific embodiments have been described in detail in the foregoing detailed description and illustrated in the accompanying drawings, those with ordinary skill in the art will appreciate that various modifications and alternatives to those details could be developed in light of the overall teachings of the disclosure. Accordingly, the particular arrangements disclosed are meant to be illustrative only and not limiting as to the scope of the invention, which is to be given the full breadth of the appended claims and any and all equivalents thereof.