Methods to assist a driver of a vehicle in avoiding accidents with pedestrians

Abstract

Method for assisting a driver of a vehicle to avoid accidents with pedestrians using an environment sensor, comprising the following steps: a) Determining an upper limiting speed of an object in order to just leave the lane of one's own vehicle at a given distance x before the own vehicle has covered the distance x, b) Determining a lower limiting speed of the object in order to just reach the driving lane of the own vehicle at a given distance x before the own vehicle has covered the distance x, c) Determining a probability distribution of possible velocities of the object, d) Calculation of the collision probability based on the values ​​determined in steps a) to c) and e) Output of the collision probability to a warning or intervention function.

Description

[0001] The invention relates to a method for assisting a driver of a vehicle to avoid accidents with pedestrians using at least one environmental sensor.

[0002] WO 2009 / 141092 A1 discloses a driver assistance system for avoiding collisions between a vehicle and pedestrians, comprising a camera and / or beam sensor. When an object is detected moving at a certain speed on a pedestrian crossing, the object is classified as a pedestrian with a probability high enough to issue a warning to the driver and prevent a potential collision. This method is therefore dependent on the accuracy of determining the speed of an object that could be a pedestrian.

[0003] WO 2011 / 141 018 A2 discloses a method for assisting a vehicle driver, whereby an environmental sensor detects a lane in front of the vehicle to determine a driving lane for a road vehicle in a complex traffic situation. The driving lane is determined based on detected lane markings and raised road edges.

[0004] DE 10 2013 212 473 A1 discloses a method for the autonomous activation of a system to assist a driver of a vehicle in the event of an impending collision. The autonomous activation of the system is based on an assessment of the criticality of the situation.

[0005] The object of the invention is to provide an improved method to assist a driver of a vehicle in avoiding accidents with pedestrians.

[0006] A starting point of the invention is the realization that it is difficult to measure or accurately estimate the speed of an object, which could be a pedestrian, traveling perpendicular to the direction of travel of the vehicle. Different types of environmental sensors inherently offer different object detection capabilities, speed resolutions, and errors.

[0007] A basic idea of ​​the invention lies in the assessment of the criticality of a situation involving an object that is at least potentially a pedestrian.

[0008] One aspect of the invention relates to the further development of a generally known emergency brake assist (EBA) system that brakes for pedestrians to protect them. Another aspect of the invention relates to determining the probability of a collision with a pedestrian.

[0009] Another aspect of the invention relates to a method that can be used independently of the sensor type of the environmental sensor.

[0010] An inventive method for assisting a driver of a vehicle to avoid accidents with pedestrians using an environment sensor comprises the following steps: a) Determining an upper limiting velocity v_max of an object in order to just leave the lane of the own vehicle at a given distance x before the own vehicle has covered the distance x, b) Determining a lower limiting speed v_min of the object in order to just reach the driving lane of the own vehicle at a given distance x before the own vehicle has covered the distance x, c) Determining a probability distribution p of possible velocities v of the object, d) Calculation of the collision probability based on the values ​​determined in steps a) to c) and e) Output of the collision probability to a warning or intervention function of the vehicle.

[0011] The vehicle's environmental sensor can be a camera, in particular a stereo camera, a lidar or radar sensor, a surround-view camera system, a laser scanner, a photon mixing detector (PMD), or any other sensor capable of capturing the vehicle's surroundings or providing information about them. A telematics device that receives environmental data from other vehicles (vehicle-to-vehicle) or infrastructure (vehicle-to-X) could also serve as an environmental sensor in this sense. Advantageously, multiple environmental sensors can be used instead of just one, enabling a more comprehensive understanding of the vehicle's surroundings.

[0012] Objects detected by the environmental sensor can be pre-classified, e.g., as potential pedestrians, if the sensor data does not rule out, or strongly suggests, that the object is a pedestrian. This pre-classification can take into account characteristic pedestrian features such as size, speed, and anatomical features.

[0013] Advantageously, the environmental sensor detects the spatial arrangement of the object in relation to the vehicle itself or to the environmental sensor permanently mounted in the vehicle. Preferably, kinematic parameters of the objects detected by the environmental sensor are determined, such as distance, speed, and acceleration, as well as the variance of these parameters.

[0014] The so-called "travel path" of a vehicle is generally the two-dimensional area that indicates the territory the vehicle traverses and is traversed during its journey. Advantageously, data from at least one environmental sensor, the vehicle geometry, and / or one or more vehicle sensors are considered to determine the travel path. These sensors provide information such as the vehicle's own speed, vehicle acceleration, steering angle, yaw angle, and yaw rate. Environmental data from at least one sensor can also be incorporated into the prediction of the travel path, such as lane markings and boundaries, obstacles on the road, and similar features.

[0015] A probability distribution of possible speeds of the object or pedestrian can be determined in the form of a frequency distribution, probability density function, or continuous probability distribution function. The frequency distribution or probability density function expresses, for n discrete speeds, the probability with which the pedestrian moves at the respective speed.

[0016] In the simplest case, a typical pedestrian speed can be assumed as an average value if the direction of movement of the pedestrian can be determined, e.g. from his positioning.

[0017] Advantageously, the probability distribution can take into account past positions of the pedestrian detected by the environmental sensor, as these provide information about the pedestrian's previous speeds. It is therefore essential that the determination of the collision probability is not based on the pedestrian's current speed, but rather on a probability distribution of the pedestrian's speeds, advantageously consisting of at least five different discrete speed values.

[0018] Another aspect of the invention relates to a corresponding device for assisting a driver of a vehicle to avoid accidents with pedestrians.

[0019] Advantageous embodiments of the invention are the subject of the dependent claims and the following description, as well as the figures. Exemplary embodiments of the invention are described below and explained in more detail with reference to the figures.

[0020] In a preferred embodiment, the probability that a pedestrian will be in front of the vehicle at the same x-distance at a given time is determined in several steps: Step 1: Calculation of the pedestrian's required speeds in the direction of the ego trajectory to just reach or just leave the driving lane of one's own or ego vehicle. Step 2: Calculation of a probability density over possible speeds of the pedestrian in the direction of the ego trajectory. Step 3: Integration of the probability density between the two velocities as determined in step 1.

[0021] The result of the integration is the final result of the calculation and expresses the probability of a collision with a pedestrian.

[0022] This method improves existing emergency braking assistance systems by enabling them to calculate the criticality of a pedestrian situation independently of sensors. It is a prerequisite for triggering a warning, braking, or evasive maneuver.

[0023] A particular advantage of this method is that even poorly measured lateral object speeds can be compensated for. This is achieved through a probabilistic estimation of the pedestrian's position at the time of impact, without requiring an explicit estimate of the pedestrian's speed.

[0024] State-of-the-art methods use, for example, a prediction of the pedestrian position by assuming that the current kinematics remain constant.

[0025] The probabilistic approach of the method demonstrates high robustness in everyday situations for various sensor types.

[0026] Fig. Figure 1 schematically shows a vehicle (1) with an environmental sensor, e.g., a camera, lidar, or radar sensor, driving on a road (2). A pedestrian (3) is located at the lower edge of the road (2). The vehicle's environmental sensor (1) determines the position of the pedestrian (2), i.e., the distance in the longitudinal direction (x-direction) between the front of the vehicle and the pedestrian (3), as well as the distance in the transverse direction (y-direction).

[0027] Optionally, the environmental sensors determine movement data of the pedestrian (3), i.e., his speed in the x and y directions, preferably at successive times. The acceleration of the pedestrian (3) can also optionally be determined from data of the environmental sensor.

[0028] The travel path (4) of the ego vehicle (1) can be determined or estimated taking into account the actual position, the actual speed v_ego, the width, the yaw angle and / or the yaw rate of the own vehicle. Details for step 1:

[0029] The necessary speeds of the pedestrian (3) at which a collision can occur are calculated as follows: The distance d_min to be covered by the pedestrian (3) is determined between the foremost point of the pedestrian (3) and the nearest point of the driving tube (4) of the vehicle (1). Similarly, the distance d_max to be covered is determined between the rearmost point of the pedestrian (3') and the furthest point of the driving tube (4). Fig. Figure 1 shows the pedestrian (3') at the corresponding position above the driving tube (4) as an example.

[0030] The measured distances are divided by the time the vehicle (1) needs to travel the distance to the pedestrian (3) in a longitudinal direction: v_min=d_min / (d_long / v_ego)v_max=d_max / (d_long / v_ego) where v_min and v_max are the pedestrian speeds to be calculated, d_min and d_max are the distances as described in the previous paragraph and in Fig. Figure 1 shows where d_long is the distance of the ego vehicle to the pedestrian in the longitudinal direction and v_ego is the speed of the ego vehicle in the longitudinal direction.

[0031] As long as the pedestrian (3) is moving slower than the minimum speed or faster than the maximum speed, there is no danger of a collision between the vehicle (1) and the pedestrian (3) and no measure to assist the driver, such as a driver warning or automatic emergency braking, is required. Details for step 2:

[0032] The probability density expresses, for n discrete speeds, the probability that the pedestrian (3) moves at the respective speed.

[0033] In the simplest case, an average speed of 1.39 m / s can be assumed, which corresponds to a typical pedestrian speed of 5 km / h.

[0034] The probability density can be determined more precisely by performing a linear curve fitting on the last m lateral distances of the pedestrian (see Fig. 2) Five lateral distances (m=5) are shown there, corresponding to the five most recent object positions detected by the environmental sensor. This fitting is performed for each of the n discrete velocities. The respective velocity determines the slope of the line. When searching for the line with the smallest deviation error, the support points located in the recent past are given greater weight. Fig. Figure 2 shows the fitting line for the discrete velocity value v = -3.5 m / s, which, with this slope, best represents the m lateral distances Y_Dist under the given conditions. However, the error of this fitting line is relatively high, resulting in a low probability for v = -3.5 m / s. In particular, a quantity proportional to the inverse of the deviation error can be entered as a probability into the probability density function for this discrete velocity value. The area of ​​the probability density function is advantageously normalized to one.

[0035] The variances in the detection accuracy of the detected y-object positions can be taken into account during curve fitting. Measurements with lower variance have a greater influence on the determination of the curve fitting error. Thus, the variances affect the probability density of the possible object velocities. In this way, measurement uncertainties can be considered when determining the probability distribution of possible object velocities.

[0036] To account for various measurement inaccuracies of the sensors used and to maximize flexibility, a Gaussian kernel can be used as a probability density in addition to curve fitting. This kernel has the measured speed of the pedestrian in the direction of the ego trajectory as its mean and the measured speed variance as its variance.

[0037] Both probability densities are added together with a weighting. The weight of the Gaussian probability function is determined from the quality of the detected object.

[0038] In Fig. Figure 2 shows an example of the pedestrian's lateral distance Y_Dist plotted against time t. The values ​​of the last five time steps (circles) are shown, along with a fitted straight line (dashed line) for a specific speed. It can be seen that the straight line is closer to the points measured later.

[0039] In Fig. Figure 3 shows the probability density of pedestrian speeds (solid curve). This indicates the probability that a pedestrian is moving at a specific speed or within a specific speed range. Details for step 3:

[0040] In the third step, the probability density between the two speeds v_min and v_max, as determined in step 1, is integrated. This yields the probability of a collision with a pedestrian.

[0041] The speeds v_min and v_max determined in step 1 define the speed range over which integration is performed. The probability density determined in step 2 for discrete speed values ​​is shown as a solid curve p(v). This continuous probability distribution can be interpolated from the discrete probability density. The hatched area shows the integration result and thus the probability of a collision between the vehicle (1) and the pedestrian (3). Depending on this integration result, targeted measures can be taken to avoid an impending accident with the pedestrian (3). This could involve an acoustic, visual, and / or haptic warning to the driver of the ego-vehicle (1). The brakes of the ego-vehicle (1) can be pre-filled, or braking initiated by the driver can be intensified in this situation.An automatic braking intervention (emergency braking) to avoid a collision or to reduce the vehicle's speed in the event of a collision with the pedestrian (3) can occur. An evasive maneuver can be initiated if the situation allows. The latter is also known as Emergency Steer Assist (ESA). To protect the pedestrian (3) in the event of an actual collision, further measures can be taken, such as raising the vehicle's hood (1) or activating a pedestrian airbag.

Claims

[1] Method for assisting a driver of a vehicle to avoid accidents with pedestrians using an environment sensor comprising the steps: a) Determining an upper limiting speed of an object in order to just leave the lane of one's own vehicle at a given distance x before the own vehicle has covered the distance x, b) Determining a lower limiting speed of the object in order to just reach the driving lane of the own vehicle at a given distance x before the own vehicle has covered the distance x, c) Determining a probability distribution of possible velocities of the object, d) Calculation of the collision probability based on the values ​​determined in steps a) to c) and e) Output of the collision probability to a warning or intervention function. [2] Method according to claim 1, wherein environmental sensor data on the previous speed of the object are incorporated into the probability distribution of possible speeds of the object. [3] Method according to claim 1 or 2, wherein the variances of the position and / or velocity data of the object determined by the environmental sensor are taken into account when determining the probability distribution of possible velocities of the object. [4] Method according to one of the preceding claims, wherein the calculation of the collision probability comprises an integration of the probability distribution of possible velocities of the object from the lower to the upper limiting velocity of the object. [5] Method according to one of claims 2 to 4, wherein a probability distribution of possible velocities of the object is determined from several past object positions detected by the environment sensor. [6] Method according to claim 5, wherein the object positions that were detected closer in time to the current time are given greater consideration in determining the probability distribution of possible velocities of the object than object positions further in the past. [7] Method according to claim 5 or 6, wherein the determination of the probability distribution of possible velocities of the object comprises a linear curve fitting over the detected y-object positions. [8] Method according to claim 7, wherein the linear curve fitting is carried out such that for several discrete velocity values ​​of the object in the y-direction a regression line with a slope corresponding to the discrete velocity value is determined to the detected y-object positions and the error of the regression line from the detected y-object positions is determined for the discrete velocity values. [9] Method according to claim 8, wherein the probability distribution of possible velocities of the object is determined such that a high error for the regression line of a discrete velocity value leads to a low probability for that velocity value. [10] Device for assisting a driver of a vehicle to avoid accidents with pedestrians comprising at least one environmental sensor designed to detect the vehicle's surroundings; an investigative unit, trained to a) Determining an upper limiting speed of an object in order to just leave the lane of one's own vehicle at a given distance x before the own vehicle has covered the distance x, b) Determining a lower limiting speed of the object in order to just reach the driving lane of the own vehicle at a given distance x before the own vehicle has covered the distance x, c) Determining a probability distribution over possible velocities of the object; a calculation unit, trained for d) Calculation of the probability of collision based on the values ​​determined by the investigation unit; and an output unit, trained to e) Output of the collision probability to a warning or intervention device.

Images