TW201704067A - Collision avoidance method, computer program product for said collision avoidance method and collision avoidance system - Google Patents

Abstract

There is provided a collision avoidance method for vehicles, the method comprising the steps of: detecting at one or more sensors mounted on a vehicle in use the position of an object relative to the vehicle at two or more time instances; receiving, at the input of a processor having at least one input and at least one output, the detected positions of the object from the one or more sensors; approximating the past trajectory of the object relative to the vehicle based on the detected positions; predicting the future position of the object relative to the vehicle using the approximated past trajectory; estimating the likelihood of collision of the vehicle with the object based on the predicted future position of the object relative to the vehicle; determining whether the likelihood of collision is over a predetermined threshold; and if the likelihood of collision is over the predetermined threshold, outputting an alarm signal from the output of the processor. There is also provided a computer program product for a collision avoidance method. There is further provided a collision avoidance system for vehicles for performing the above method steps.

Description

Anti-collision method, computer program product and anti-collision system for realizing the anti-collision method

The present invention relates to a collision avoidance method, a computer program product for implementing the collision avoidance method, and a collision avoidance system, and more particularly to a method for avoiding collision between a vehicle and a vulnerable road user, including a cyclist and a pedestrian. Method, computer program product and system. It offers more significant advantages over current technology in terms of accuracy and effectiveness of detection and prevention of potential collisions.

Collision avoidance systems are developed by automakers to actively reduce the likelihood of potential collisions and reduce the severity of road traffic accidents. These systems typically use radar, laser, or video cameras in front of and behind the vehicle to detect impending collisions and then provide an alert or perform an autonomous evasive technique such as automatic braking or steering to avoid impending collisions. 1 shows a conventional collision avoidance system 10 including a potential risk assessment system 11 that is capable of receiving information from a condition recognition system 12 (including forward and backward CCD cameras, radar or laser light) and vehicle sensors (including Input of the wheel rate sensor and the acceleration sensor), and then outputting the control signal to an early warning system 13 (including a buzzer, a head up display, and a stop light) and a vehicle brake control system 15 (including a brake actuator and Sub-throttle accelerator).

Some known collision avoidance systems employ a multi-stage collision avoidance mechanism that applies different levels of collision avoidance techniques depending on the likelihood or severity of the predicted collision; if the likelihood of a collision is classified as low, The collision system can activate a warning light or another warning signal in the vehicle to alert the driver that if the possibility of a collision is classified as high, the collision avoidance system can be used in the automatic braking of the vehicle.

Vehicles with anti-collision systems are usually equipped with adaptive cruise control, which uses radar or laser light to automatically adjust the speed of the vehicle to maintain a safe distance from the preceding vehicle. Sometimes, cooperative adaptive cruise The use of control systems allows most road users to exchange information with each other, allowing road users on a single line to be equidistant from one another, further reducing the likelihood of collisions.

However, despite various attempts to reduce the number of road collisions, traffic accidents are still a major cause of injury and death. Although the head-off system has a high market share in the past, many collisions actually occur at busy intersections in urban areas, where there is usually more activity and the visibility of drivers in all corners may be reduced, but Current collision avoidance systems are not necessarily suitable for detecting and reacting to these types of potential collisions.

The inventors have found that there is a need to propose improved collision avoidance systems and methods to meet the problem of collision detection at intersections.

Accordingly, an embodiment of the present invention provides a vehicle collision avoidance method, the method comprising: detecting one at two or more time points by one or more sensors installed in the vehicle in use Position of the object relative to the vehicle; receiving an object detected by the one or more sensors with an input having at least one input and at least one output; estimating based on the detected position Predicting the previous trajectory of the object relative to the vehicle; using the estimated previous trajectory to predict the future position of the object relative to the vehicle; estimating the collision of the object with the object based on the predicted future position of the object relative to the vehicle Possibility; determining whether the likelihood of collision exceeds a predetermined threshold; and, if the probability of collision exceeds a predetermined threshold, outputting an early warning signal by the output of the processor.

In particular, the present invention focuses on vehicle cornering (especially the inside lane of a road) and side collisions of objects (or near its sides) that are close to the side of the vehicle within the radius of gyration of the vehicle. The position of the object relative to the vehicle is defined by the distance between the two and its orientation data.

The output of the sensor, ie the detected distance, can be an analog or digital voltage output. The detection distances from at least two of the plurality of sensors may be filtered first and then transmitted to a processor for reception, such as a real-time controller, to perform additional method steps to provide a more immediate Collision avoidance system.

By detecting the relative position of the object at at least two points in time, the relative velocity of the object can be calculated. By detecting at two points in time, it can be assumed that the relative velocity of the object is constant. With three detections, it is possible to assume that the acceleration of the object relative to the vehicle is constant. More detection, such as 10 to 20 detection events, allows the step of estimating the previous movement of the object to be more precise. Preferably, the position of the object is detected at 10 to 20 o'clock, and the sensor is preferably detected at a reading rate of at least 7.5 Hz at 15 o'clock.

Based on the detected position, a previous movement trajectory of the object relative to the vehicle is estimated, and then the estimated previous movement trajectory is used to predict the future position of the object relative to the vehicle. The active collision avoidance method (relative to the passive method) that takes into account the previous movement trajectory of the object relative to the vehicle helps to more accurately predict the future position of the object. When estimating the future position of the object relative to the vehicle, the processor then determines whether the likelihood of collision of the vehicle with the object exceeds a predetermined threshold, and if the likelihood of collision exceeds a predetermined threshold, the output of the processor outputs an early warning signal .

Using at least two detection events, you can determine whether the object is moving or fixed. In the case of a fixed object whose object is a lamp post or a mail box, the processor recognizes that the speed of the object relative to the vehicle is negative (or the object may no longer be detected, for example, after the first detection event), It may result in an estimated previous trajectory of the object and a predicted position relative to the future of the vehicle, causing the likelihood of collision to be less than a predetermined threshold. In one example, the object is a moving object of a pedestrian, other vehicle, or the like that travels backward relative to the vehicle, and the processor recognizes that the speed of the object relative to the vehicle is a negative value, and the determined collision probability is also less than a predetermined threshold. In one example, the object is a moving object of a pedestrian, other vehicle, or the like that travels forward relative to the vehicle, and the processor can identify that the object is traveling at a negative or positive value relative to the vehicle. In the example, it is preferred to use more than two detection events to retrieve more information about the object's previous movement trajectory relative to the vehicle, for example, using three detection events to identify the object. Acceleration.

Preferably, the one or more sensors are arranged along a horizontal axis of one side of the vehicle in use. In some embodiments, a plurality of sensors (eg, 12 sensors) are preferably used such that one side of the vehicle can be fully covered along the horizontal axis of the vehicle use side.

The one or more sensors may be disposed substantially along a horizontal axis of a top edge of one side of the vehicle in use, the sensors being respectively downwardly facing the ground or toward the rear of the vehicle, or placed at a distance from the ground 0.5 to 1.5 meters in height and pointing to the back of the vehicle. (This configuration is particularly helpful where the one or more sensors include a position sensor like a camera). Alternatively, preferably, the one or more sensors may be arranged along a horizontal axis, preferably at a height of 0.5 to 1.5 meters from the ground, and the respective sensors may point in the same direction Preferably, the one or more sensors may be configured to be perpendicular to each other and directed horizontally outward from the side of the vehicle to the outside of the vehicle. (This configuration is particularly useful where the one or more sensors include at least two distance sensors such as ultrasonic sensors or the like).

The one or more sensors may comprise at least two distance sensors, preferably at least two ultrasonic sensors. In a preferred embodiment, it is possible to use 10 to 20 ultrasonic sensors that can be placed along an axis of the vehicle at an interval of about 1 meter, and in a particularly preferred example In this case, 12 ultrasonic sensors can be used such that they are arranged substantially along and horizontally covering the horizontal axis of the vehicle. Sensors configured in this manner are particularly useful for detecting objects beside the vehicle.

The one or more sensors may be calibrated to detect their distance to the object to be detected. The sensor can detect the distance between the vehicle and the object to be detected, respectively. The azimuth information can be further determined using at least two sensors. How to implement this embodiment will be described in detail with reference to the accompanying drawings.

In an example in which at least two sensors are used, the sensor can separately detect the distance between the vehicle and the object to be detected at a time point. Preferably, the sensor used is an ultrasonic sensor. Ultrasonic sensors can be selected for the proposed method embodiments for at least the following reasons: i. Ultrasonic sensors can accurately and reliably measure moving targets; ii. Output at speeds of at least 7.5 per second Detection is possible; iii. Suitable for outdoor and various weather conditions; iv. With internal temperature compensation; v. Easy to use a real-time controller to process the sensor output; vi. Ultrasonic sensor is not only lightweight Cost-effective; and vii. A detection distance of approximately 0.3 - 5 meters and a good measurement resolution of approximately 5 mm.

The one or more sensors may alternatively or additionally include at least one position sensor capable of determining a distance and an azimuth between the sensor and the object, for example, a visual sensor ( Or camera) (allows infrared or video wavelengths to operate), radar detectors or lidar detectors (or laser scanners). The captured image can then be processed by the processor to obtain the position of the object relative to the vehicle, including distance and orientation information from the vehicle.

In so-called "fused" systems, which use different types of combined sensors, the outputs of the various sensors are integrated or fused to provide the position of the object relative to the vehicle at a particular point in time. More accurate detection. In a fused system, one or more proximity sensors, which are binary detectors, can detect any presence of an object that is not actually in contact within the rated range. These sensors can be used in conjunction with the one or more sensors to increase the accuracy of the system. A plurality of proximity sensors and one or more cameras may also be included in the fused system. The proximity sensor is used to record the distance between the object and the vehicle, and the camera helps the object to recognize and set the detection of the proximity sensor in different groups. Cameras can be used in conjunction with image processing software to distinguish between different types of objects in adjacent vehicles (eg, street furniture, parked cars, and a group of cyclists) that can be "fused" with data from proximity sensors. To get a more powerful and reliable way to determine the action of each cyclist.

Preferably, the method further includes filtering out irrelevant detections to obtain a series of useful detection steps. This can be referred to, for example, "ID Screening Method".

Preferably, the step of estimating the current movement trajectory of the object can be implemented according to an optimization method. Preferably, the optimization method, in a preferred case, minimizes the cost function of the object relative to the previous movement trajectory of the vehicle in order to select the most likely associated with the one or more sensors at various points in time. Azimuth sequence. The cost function is designed to ensure that the trajectory is smooth and can be used, for example, rms longitudinal acceleration or root mean square longitudinal velocity. Preferably, the optimization method is a quadratic programming method that provides a method of optimizing a series of detected object positions (distance and azimuth) at a number of points in time to produce a smooth trajectory. It can be assumed that the longitudinal velocity and/or acceleration progression is constant.

The optimization method may additionally or additionally include analyzing the series of detections to establish inequality constraints. This is especially useful when using at least two distance sensors (eg two ultrasonic sensors). When the one or more sensors include at least two distance sensors, preferably the optimization method may include using the distance measured by the two adjacent sensors to the object The position is triangulated to set equality constraints. Ultrasonic sensors only output distance information, so using an ultrasonic sensor does not give the position information needed to determine the position associated with the sensor. In order to determine the position information, the azimuth is unknown in each formula. When adjacent sensors have overlapping fields of view, triangulation using two adjacent sensors can provide additional orientation information. This technique can be used to fix the position of the object relative to the vehicle at a point in time, and to derive the azimuth angles of the two sensors, which are used as equality constraints in quadratic programming.

Triangulation between pairs of adjacent sensors can be performed. Triangulation provides two azimuths for two sensors, and any azimuth can be used to indicate the estimated position of the object to be detected. The sensor near the front side of the vehicle is called the front sensor, and the other sensor in the triangle is called the tail sensor. The step of predicting the movement trajectory of the object relative to the vehicle is performed according to the distance detected by one of the sensors and the azimuth of the sensor. However, preferably, two sensings should be used. The distance and azimuth of the device are used to achieve triangulation, and the detection distance of the sensors is shorter than the distance to be detected. For example, if the detection distance of the tail sensor is greater than the detection distance of the front sensor, in the preferred case, the detection distance and the azimuth of the front sensor are used to predict the object. Future movement trajectory. In another example, if the detection distance of the front sensor is greater than the detection distance of the tail sensor, in the preferred case, the detection distance and azimuth of the tail sensor are used. Predicting the future trajectory of the object. In yet another example, if the tail sensor and the front sensor have the same detection distance, the tail sensor or the front sensor may be selected depending on the previous detection.

The collision avoidance method may further comprise the step of smoothing the estimated previous movement trajectory of the object using a Kalman filter. The movement trajectory of the object to be detected can be smoothed according to the motion pattern of the object relative to the motion of the vehicle.

The step of predicting the future trajectory of the object may be performed under the assumption that the acceleration and/or yaw rate of the object relative to the vehicle is constant.

The method according to an embodiment of the present invention is applicable to various vehicles, particularly trucks, buses, coaches, trams, and automobiles. This method is believed to be particularly suitable for heavy goods vehicles, especially non-articulated heavy goods vehicles. In urban areas, these vehicles share roads with other potentially vulnerable road users such as cyclists and motorcyclists, which pose significant safety issues. Vulnerable road users, such as those who share roads with heavy goods vehicles, including cyclists, motorcyclists, and pedestrians. This method is believed to be particularly useful for detecting two-wheeled vehicles such as bicyclists and motorcyclists, as well as pedestrians.

It has been found that when the vehicle is a heavy-duty truck and the object to be detected is a two-wheeled motor vehicle, most accidents occur on the side. One of the factors is that side collisions of heavy goods vehicles, especially non-articulated heavy goods vehicles, are often caused by the large size of these vehicles, which makes the driving of such vehicles have more blind spots when driving, which is particularly problematic. Therefore, if the bicyclist is in the vicinity of a heavy goods vehicle, such as a traffic light at an intersection, the heavy truck driver may not notice the bicycle rider at all. In addition, in order to make a smooth turn, heavy goods vehicles usually need to pull in the opposite direction (for example: pull to the right to turn left) to ensure that the angle of rotation is sufficient. This may cause the cyclist to mistake the heavy-duty truck for not turning towards them (and continuing straight or turning away from the cyclist), which may cause the cyclist to move forward to the blind spot area of the heavy truck driver. Then, the front wheel/corner of the vehicle's rotation may knock the biker down, and the rear wheel of the vehicle may crush the biker. Accordingly, a preferred method of the present invention is that the vehicle can be a heavy goods vehicle and/or the object being detected is a two-wheeled motor vehicle, such as a bicycle.

Preferably, the early warning signal output by the processor may include a video and/or audio warning signal, and the method may further include the step of: sending an early warning signal to a video/audio if the probability of a collision exceeds a predetermined threshold An alarm to activate the alarm. Additionally, or alternatively, the warning signal may include a brake signal, and the method may further include the step of: transmitting the brake signal to a vehicle brake system to activate the brake of the vehicle if the probability of the collision exceeds a predetermined threshold, and / or the warning signal may comprise a turn signal, the method may further comprise the step of transmitting the steering signal to a control system of the vehicle to change the steering angle of the vehicle if the likelihood of a collision exceeds a predetermined threshold.

When activated by an early warning signal, the video/audio alarm includes a video/audio warning signal to alert the vehicle driver to a potential collision. When activated by an early warning signal, the vehicle brake system includes a brake signal for automatically or immediately automatically applying a vehicle brake without any driver input. The vehicle control system, driven by the warning signal, will automatically change the steering angle of the vehicle without any driver input. In the preferred case, the steering angle of the vehicle may be changed to be in the opposite direction to the sensor.

The steps of transmitting a video and/or audio alert signal, transmitting a brake signal, and transmitting a turn signal may be performed simultaneously or in any order. That is, a multi-stage collision avoidance method depending on the estimated likelihood of collision is provided. Preferably, transmitting a video and/or audio warning signal indicates that the first predetermined threshold of the likelihood of collision is reached in the first phase of the method, and transmitting a turn signal indicates the second phase of the method. A second predetermined threshold for the likelihood of a collision is reached, and a braking signal is transmitted indicating a third predetermined threshold in the third phase of the method when the likelihood of a collision is reached. Alternatively, the steering angle and braking signal may be applied simultaneously with the second phase of the method substantially after the video/audio warning signal is transmitted. The advantage of sending a video/audio warning signal as a first stage is that the driver of the vehicle may have the opportunity to avoid collisions without the aid of any collision avoidance system. This can be avoided by relying on automatic braking or the transmission of a turn signal. However, if a higher predetermined threshold at which a collision may occur is reached, the method further includes transmitting a braking and/or steering signal to the vehicle brake system and/or control system.

In accordance with an embodiment of the present invention, the method includes transmitting only an early warning signal to a video/audio alarm, and the collision avoidance system capable of performing the above method can be isolated from a control system (e.g., a vehicle brake system or steering system). In another aspect, the method of an embodiment of the present invention includes separately transmitting a brake signal and/or a steering signal to a vehicle brake system and/or control system (or an early warning signal sent to a video/audio alarm). The collision avoidance system implementing the above method can be coupled to the vehicle brake system and/or control system.

According to another embodiment of the present invention, a computer program product for a collision avoidance method is provided, the computer program product comprising: a storage device including instructions executable by a processor to cause the processor to execute all of the above Or any method step.

According to still another embodiment of the present invention, a vehicle collision avoidance system is provided, the system comprising: one or more sensors mounted on the vehicle in use for detecting relative objects in two or more time points At a location of the vehicle; and a processor having at least one input and at least one output, the processor configured to: utilize the input thereof to receive the plurality of sensors detected by the one or more sensors Positioning; estimating a previous movement trajectory of the object relative to the vehicle based on the detected positions; using the estimated previous movement trajectory to predict a future position of the object relative to the vehicle; according to the predicted object relative to the vehicle The future location estimates the likelihood of collision of the vehicle with the object; determines if the likelihood of the collision exceeds a predetermined threshold; and outputs an early warning signal from the output of the processor if the likelihood of the collision is above the predetermined threshold. It should be understood that the various system features, methods, and advantages described above in accordance with an embodiment of the present invention are further applicable to the systems of the following embodiments.

The sensors in the collision avoidance system are respectively installed along the horizontal axis of one side of the vehicle, which is particularly useful in detecting and preventing the above-mentioned side collision or near collision. In some preferred embodiments, the sensor is an ultrasonic sensor, and the one or more sensors are respectively disposed at a height of 0.5 to 1.5 meters from the ground, and (or) the system is Two or more sensors are included, and the distance between two adjacent sensors is 0.5 to 1 meter, preferably about 0.8 meters. With such a configuration, it is believed that the collision avoidance system is particularly suitable for vehicles used in heavy goods vehicles, and the object to be detected is a two-wheeled motor vehicle or pedestrian.

Preferably, the system further includes a guide cone for at least one or each of the sensors. In one example, the guiding cone can be a plastic cup, which reduces the detection of secondary reflection by reducing the detection beam of one sensor, thereby improving the stability of detection.

Preferably, the system may also include a video and/or audio warning signal and may include an alarm. The processor can also be configured to send an early warning signal to the alarm to activate the alarm when a likelihood of collision exceeds a predetermined threshold. Additionally, or alternatively, the early warning signal may include a brake signal, and the processor may be further configured to transmit an early warning signal to a vehicle brake system to clamp the vehicle when the probability of collision exceeds a predetermined threshold, and/or The early warning signal can include a turn signal, and the processor can be further configured to transmit an early warning signal to a vehicle control system to change a steering angle of the vehicle when a likelihood of collision exceeds a predetermined threshold.

The problem with side collisions with heavy goods vehicles is that these vehicles are usually large and have more blind spots. Despite the extra mirrors, there are still many parts that hinder driving, and the mirror itself may hinder and/or distract the driver. Figure 2 is a plan view showing the left-hand side area that the driver can see from the ground: rear window area 21; wide-angle rear view mirror area 22; side close to mirror area 23; plane rear view mirror area 24; near side window area 25; a glass zone 27; and a front projection mirror zone 28. Areas 26, 29 are areas that cannot be seen directly or indirectly by driving. It should be understood that similar illustrations can be obtained and used for other elevational views to achieve a three-dimensional map of the visible and invisible areas of the driver of the complete heavy goods vehicle. Also, if the bicyclist is in the vicinity of a heavy goods vehicle, such as a traffic light at an intersection, the heavy truck driver may not notice the bicycle rider at all.

Furthermore, in order to make a smooth turn, the heavy goods vehicle 30 (shown in Figure 3A) typically needs to be pulled in the opposite direction (e.g., pulled to the right to turn left) to ensure that its angle of rotation is sufficient, as shown in Figure 3B. This may cause the cyclist to mistake the heavy-duty truck for not turning towards them (and continuing straight or turning away from the cyclist), which may cause the cyclist to move forward to the blind spot area of the heavy truck driver. Then, the front wheel/corner of the vehicle's rotation may knock the biker down, and the rear wheel of the vehicle may crush the biker.

4 is a block diagram showing a real-time collision avoidance system in accordance with an embodiment of the present invention. Compared to prior art systems, the system of the present invention provides a solution to the problems mentioned in Figure 3 above in a simpler and more efficient manner. The action of the object to be detected (in this case a cyclist) and the vehicle using the global coordinate system (in this case HGV) are converted into the relative motion of the object to be detected with respect to the vehicle in the coordinate system. Due to the summation of the vector of acceleration and yaw rate, the assumption of constant acceleration and constant yaw rate can be used in the coordinate system of a vehicle as in the global coordinate system.

Representing the relative motion of the object to be detected relative to the vehicle provides the following advantages: i. Eliminates the need for accurate vehicle movement data such as acceleration and position, making the system easier and saving the computation time required by the real-time system; and ii. The collision avoidance system requires less hardware and removes the expensive vehicle motion measurement system (GPS + IMU) from the system, making it more commercially attractive.

Accordingly, the real-time collision avoidance system 40 of FIG. 4 will perform the following tasks: i. processing the output from the ultrasonic sensor to estimate the position of the knight (S41); ii. estimating the time to avoid the potential collision (S42); Judging the position of the knight relative to the heavy goods vehicle (HGV) based on the detected distance and predicting the knight's future movement trajectory (S43); iv. estimating the collision possibility (S44); and v. if necessary, by the processor The output sends an early warning signal, which may be a braking command sent to the vehicle brake system (S45).

Figure 5 shows how the system 50A, 50B can be used to estimate the position of the bicyclist to avoid collisions in accordance with an embodiment of the present invention. In this example, a plurality of ultrasonic sensors are used, and the ultrasonic sensors output the objects 53, 56 to be detected (ie, the object in this case is a bicyclist) and the sensor. The distance d between. However, only with distance information, it is impossible to clearly indicate the exact position of the objects 53, 56 because the azimuth between the sensor and the object is unknown. Accordingly, the bicycle rider can be anywhere in an arc of the same radius, as shown in Figure 5A. This is called location uncertainty. In order to establish the trajectory of the cyclist relative to the vehicle, it is necessary to obtain position information to remove any positional uncertainty.

As shown in Figure 5B, a coordinate system can be provided to define the position of the ultrasonic sensor in the vehicle and to indicate the position of the bicyclist. The origin can be set at the midpoint of the front edge of the vehicle, the x-axis is directed upward toward the longitudinal axis of the vehicle, and the y-axis is perpendicular to the x-axis and pointing to the left.

The flow chart S60 shown in Figure 6 shows how to estimate the position of the bicyclist in real time. Each ultrasonic sensor outputs the detected distance and its sensor ID at each time step. In step S61, a fixed size matrix is created which contains the sensor ID and the distance detected in the continuous time step. This matrix can be called a detection matrix, and there can be multiple detections in each time step because: 1 multiple objects in the detection zone; and 2 two adjacent sensors can use the same Object. It is necessary to process the detection matrix to divide the sensor ID into different groups, each group corresponding to an object. This process is called "ID grouping and tracking" (S63). At step S63, sensor detection that is irrelevant to the movement of the bicyclist is removed. Step S62 is a camera-based method, and it is possible to freely select whether or not to use the auxiliary step S63. At step S62, a particular computer vision technique is applied to locate any of the bicyclists in the field of view, and this information is sent to S63 to cause the ID of the ultrasonic sensor to be grouped. The result of S63 is sent to S64 to check if any feasible triangulation is formed between adjacent sensors among the detection groups. Next, the quadratic programming method (QP) using the optimization algorithm is used to determine the optimum detection angle θ and the corresponding position of the bicyclist (S65). The result of the step S64 is considered in step S65 to propose an equality constraint to assist in confirming the detection angle. Finally, the Kalman filter is used to smooth the cyclist's movement trajectory according to the relative movement model of the bicyclist (S66).

Some of the steps of flowchart S60 in Figure 6 will be described in further detail. The reference line 57 can be defined to represent the azimuth angles from the various sensors and from the sides of the vehicle, as shown in Figure 5B, as well as the distance lines 52, 55 from the sensor to the object. The angle between the reference line 57 and the distance line is defined as the azimuth angle θ of the object, and the clockwise rotation of θ is defined as a positive value. Therefore, the positions P _cy and P _{cx of the} bicycle rider relative to the vehicle positions P _sy , P _sx can be expressed by the following formula:

In the above formula, only the parameter θ is unknown. Therefore, before the detected distance and the position of the sensor on the vehicle are known, the actual problem is how to choose the value as θ, assuming the detection.

It is impossible to solve the formulas (1) and (2) independently. Suppose a short time _{(t 1, t 2, ...} , t n) detected in a series of distance _{(d 1, d 2, d} 3, ..., d n), we need to find the corresponding azimuth angle (θ _1, θ ₂ ,...,θ _n ) to determine the position of the knight; that is, the following formula must be used: (3) (4) where i=1, 2,...n.

Using simple first-order and second-order differentials, you can get the knight's speed and acceleration, as shown in the following formula:

Where j=2,3,...n;k=3,4,...,n. V and A do not indicate speed and acceleration, and t is the timestamp of each detection. Assuming that n samples, the lateral and longitudinal acceleration calculations of n-2 can be obtained from equations (7) and (8).

It can be reasonably assumed that the bicyclist moves laterally and longitudinally during t1~t2 with a constant acceleration. In general, the bicyclist does not move back and forth relative to the truck in a short period of time unless there is a sudden unpredictable stop.

can P _sx , and θ represent the longitudinal acceleration A _cx : Where k = 3,4 ... ,n

For each acceleration The only unknowns are θ _k-2 , θ _k-1 , and θ _k . Since the detection range of each sensor are all small, you can use a small angle, and to estimate [theta] sinθ w unsubstituted, linear thus this issue.

Assume that the speed is constant.

Assuming the longitudinal speed of the bicyclist is constant, the mathematical expression can be obtained as follows: Equivalent to the assumption:

Equation (11) can be transformed into a quadratic programming (QP) problem by finding a set of θ _i (i = 1,...n) that best produces a smooth trajectory record for the cyclist.

QP is mainly to optimize the quadratic functions of several linearly constrained variables. The variables θ ₁ θ ₂ ... θ _n-1 and θ _n can represent as a column of vectors:

The standard formula for quadratic planning can The form is expressed as follows:

Where Q is an n x n matrix, called a quadratic matrix, and L is a ­n-­ variable row vector called a linear matrix.

The following formula indicates some of the conditional restrictions that should be met:

The equality constraint (15) strictly constrains the variables of Θ, thus producing more accurate optimization results. We can use triangulation to make certain elements of Θ form some linear inequality constraints.

The equality constraint can consist of triangulation results. The values of A _eq and B _eq must be constant. In constructing matrices A _eq , B _eq , one principle to emphasize is that the value of the inequality constraint should not be equal to or greater than the value of the Θ variable, otherwise an over-constrained target may be generated and QP cannot converge into a solution.

In the case of inequality constraints (16), the individual elements of Θ must be limited to their upper and lower boundaries. These upper and lower limits come from the range of sensors expected. In addition to upper and lower limits for each element of the set Θ, the sensor ID in the same time sequence, provide estimates of the motion trend cyclists, therefore Θ set a more limited.

If the cyclist chases over the vehicle, the azimuth associated with each sensor in each sensor's detection range is changed from -ve to +ve, and vice versa when the truck exceeds the cyclist. If the bicyclist stays within the detection range of the sensor for a short period of time, it may be necessary to infer from the previous steps the tendency of the bicyclist to move relative to the truck.

If the bicyclist stays in the range of the sensor for a long time and does not know the previous movement of the bicyclist, it is safe to assume that the bicyclist is driving to the truck at a similar speed. In this case, it is unlikely that the bicycle rider is set to a conditional limit other than the upper limit and the lower limit.

Assume constant acceleration

To illustrate that when the longitudinal acceleration of the bicyclist is not zero, but the period to inspect (PTI) is constant, the object described in equation (12) applies the following format:

among them It is the actual longitudinal acceleration of the bicyclist at PTI.

Formula (17) indicates when equal At the time, J can be minimized, which is an unknown parameter that is optimized for the beginning, because there is no prior knowledge of the bicyclist's movement.

Take As a coefficient, formula (17) can be rewritten as:

In formula (18) Does not contain any azimuth, so it does not affect the result of optimization and will be ignored. It does not contain any azimuth of the second element, so it will only affect the linear matrix L, as defined by equation (14). The quadratic matrix Q will not be added influences.

Only in Can be solved when it is an unknown value. It is recommended to use a method to find The initial estimate is sampled using an acceleration of ² m/s ² to 2 m/s ² plus an answer of 0.1 m/s ² . For use Each acceleration sample can be quadraticly planned Candidate value. Will be possible Putting into equations (3) and (4) to get the position of the bicyclist in the PTI, and using the formulas (5) to (8) to obtain the acceleration. Then, calculate the standard deviation of these accelerations. Run all possible , you can get a series of standard deviations. Compare these standard deviations, select the smallest one, and then select the corresponding one. As the best estimate of the direction of the bicyclist. It is believed that of Is the best acceleration estimate in PTI.

When implemented, there may be some inaccurate detections (ultrasonic sensing is not always reflected by the same point of the cyclist) and signal loss, so it is difficult to obtain a smooth cyclist's trajectory using only the quadratic programming method.

A smoothing method is also needed to generate a smooth moving trajectory that can derive velocity and acceleration values for use as a future predicted position. The Kalman filter smoothes the results of the quadratic programming method. It is widely used in navigation, navigation, vehicle control systems, and mobile tracking. Using a model system, the Kalman filter allows a series of measurements to be observed over a period of time, including noise and other errors, to be smoothed.

Kinematics can be used to describe the action of the bicyclist. The state vector that describes the kinematics is called S _kf , which consists of four components:

Consider the position of the bicyclist in the horizontal and vertical directions ,speed And acceleration According to the information in step k, the actions at step k+1 are respectively represented by the following formulas, where dt is the time difference between the steps:

Considering equations (20) to (24), the state space formula of the Kalman filter can be expressed as follows:

The control input U _kf is the acceleration of the cyclist and can be assumed to be zero. w _kf represents Gaussian white noise in this process. Assume that these noises have a zero mean and a covariance matrix is defined as Go.

The measurement formula is as follows:

v _kf is the measurement noise, which is the Gaussian white distribution. The Kalman filter operates in two phases: time updates and measurement updates. A time update, also referred to as prediction, is a value used to predict the current state variable and the error covariance estimate to obtain a priori estimate for the next time step. The following two formulas are called time update formulas, where Z is the covariance matrix:

In the above measurement update formula, new measurements are added to the prior estimates to obtain an improved posteriori estimate. When new measurements are available, these estimates are updated based on the estimated values and the measured weighted averages, and multiple weights are assigned to the estimates with higher certainty. '^' indicates that the derived value is an estimate, and the superscript '-' indicates a priori estimate.

In the measurement update phase, the Kalman gain K is updated using equation (29). Based on the updated Kalman gain, the state S is re-estimated considering the measurement M of equation (30). At the same time, the updated covariance matrix P is as shown in equation (31).

The variable I _kf in the formula (31) is a 4 x 4 unit matrix. The detailed values of the matrices Z _kf , G _o , R _o are as follows:

Equations (27) ~ (31) are the core formulas of the Kalman filter and can be performed recursively to find the best estimate of the state vector.

Referring now to Figure 7, a method step of detecting the position of an object relative to a vehicle based on the detected distance using triangulation in one embodiment will be described. By placing the ultrasonic sensors close to each other, that is, at a distance of only 0.8 m (Δk), triangulation techniques can be used to fix the position of the bicyclist when the beams of two adjacent sensors overlap.

If the two adjacent sensors US _k and US _k+1 have overlapping detection ranges, the bicyclist can use the cosine law according to the two detected distances d _k , d _k+1 in the overlapping region. Find the azimuth of the sensor according to equations (3) and (4).

The azimuth angles θ _k and θ _{k+1 of the} sensor are as follows:

When certain physical conditions are met, the above triangulation technique can only produce real results, so the detection can be performed on adjacent sensors according to the following procedure:

Once the triangle measurement is found, it can be detected whether the estimated azimuth is valid by comparing the estimated azimuth and forming the maximum angle of the sensor field of view. Considering the detected distance, the azimuth must be limited by the field of view of the sensor.

Choosing the optimal spacing of the ultrasonic sensors will prevent general detection between three adjacent sensors. Therefore, if reasonable detection is performed with three adjacent sensors at the same time, it can be assumed that two different objects have been detected.

Triangulation involves two azimuths of two sensors, any of which can be used to indicate the position of the estimated bicyclist. The sensor near the front side of the vehicle is called the front sensor (ie US _{k of} Figure 7), and the other sensor in the triangle is called the tail sensor (ie US _{k+1 of} Figure 7) . In the secondary planning phase, each time only one detected distance and its corresponding azimuth are used. Preferably, the azimuth of the sensor closer to the object to be detected should be used. For example, if ρ _{k is} greater than ρk+1, the tail sensor is selected, and if ρ _{k+1 is} greater than ρ _k , the front sensor USk is selected. If ρ _k is equal to ρ _k+1 , the tail sensor US _k+1 or the front sensor US _k is selected to be used in accordance with the previous detection situation.

8A and 8B illustrate the effect that triangulation may have on the collision avoidance method in accordance with an embodiment of the present invention, wherein an object to be detected (in this case, a bicyclist) passes through a series of sensors (in This condition is an ultrasonic sensor). In Figure 8A, there is no overlap between the sensing geometries of the sensors, and in Figure 8B, the sensing geometries of the sensors overlap somewhat. The cyclist passes the sensor's detection range US10, US9, US8, US7 in a short period of time (t _n-7 ~ t _n ), for example 1 second; at the same detection interval (Period to Inspect, PTI) In the case, the sensors US1, US2 generate effective detection due to noise or another object at certain time steps. The real-time controller must be able to ignore the detection of the sensors US1, US2, as these detections may distort the predicted cyclist's movement trajectory.

Referring to Figure 8A, if there is no triangulation in any of the time steps within the PTI, the ID screening method can be performed in a general manner. First, any ID that provides effective detection at a time step can be selected to form the ID required for a string of PTIs. An example of a string of IDs may be: [US10 US10 US2 US9 US1 US7 US7 US7], where the string of numbers can be extracted to obtain a pure array of numbers [10 10 2 9 1 6 7 7]. Starting with each row in the table, these IDs have other combinations; for example, [10 10 2 9 8 8 7 10] may be another combination. Once all possible ID combinations are found, they can exist in a matrix (an ID storage matrix). The difference between the IDs of the sensors can be estimated and an optimal sensor ID string can be found.

Referring to FIG. 8B, if a valid triangle is found during the PTI process, these sensor IDs associated with the triangulation are stored. The ID associated with the shorter detection distance in each set of triangulation is used as the ID limit, which divides the complete ID string in the PTI into several ID segments. The IDs in each ID section are filtered by the same rules, including: i. ID jumps are not allowed; ii. Non-monotonic ID sections are not allowed; and iii. Response distance outliers are not allowed.

After using the above steps, you can find the best ID order, for example: [10 10 9 9 8 8 7 7]. This ID string, along with the associated detected distance, is passed to the next step of the azimuth optimization procedure called quadratic programming.

It is to be understood that while the foregoing embodiments of the invention have been described with respect to the preferred embodiments, the principles of the embodiments can be applied to all embodiments and embodiments of the invention, including those defined in the scope of the claims. In addition, the various functions described in the context of the various embodiments may be presented in combination with the various embodiments, and the features described in the context of the various embodiments may also be presented separately or in any suitable combination.

10‧‧‧Used collision avoidance system
11‧‧‧ Potential risk assessment system
12‧‧‧ Situation Identification System
13‧‧‧Warning system
15‧‧‧Vehicle Brake Control System
21‧‧‧ Rear window area
22‧‧‧ Wide-angle rearview mirror area
23‧‧‧Side close to the mirror area
24‧‧‧Flat rear view mirror area
25‧‧‧Close to the side window area
26, 29‧‧‧ Area
27‧‧‧windshield area
28‧‧‧Front projection area
30‧‧‧ Heavy goods vehicles
40‧‧‧ collision avoidance system
50A, 50B‧‧‧ system
53, 56‧‧‧ objects to be detected
52, 55‧‧‧ distance line
57‧‧‧ baseline

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Various preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings in which: FIG. 1 is a block diagram showing a collision avoidance system of the prior art; FIG. 2 is a view showing a specific embodiment according to the present invention. Figure 3A is a schematic view showing an example of a vehicle according to an embodiment of the present invention; Figure 3B is a plan view showing a possible movement of a vehicle according to an embodiment of the present invention; A specific embodiment shows an exemplary flow chart of the anti-collision method; FIG. 5A is a schematic diagram showing the operation of an exemplary sensor according to an embodiment of the present invention; FIG. 5B is a view showing an embodiment according to an embodiment of the present invention. FIG. 6 is a flow chart showing an exemplary method for displaying a predicted object movement trajectory according to an embodiment of the present invention; FIG. 7 is a diagram showing two example sensors for explaining according to an embodiment of the present invention; How to implement triangulation; FIG. 8A is an exemplary embodiment of the invention showing an example collision avoidance system without triangulation; And Figure 8B is an exemplary embodiment of the invention showing an example collision avoidance system using triangulation.

S41‧‧‧ Process the output from the ultrasonic sensor to estimate the position of the knight

S42‧‧‧ Estimated time to avoid potential collisions

S43‧‧‧ Judging the position of the knight relative to the heavy goods vehicle (HGV) based on the detected distance and predicting the future movement of the knight

S44‧‧‧ Estimated collision probability

S45‧‧‧ If necessary, an early warning signal is sent from the output of the processor, which may be a braking command sent to the vehicle's brake system

Claims (39)

A vehicle collision avoidance method, the method comprising the steps of: detecting a position of an object relative to the vehicle at one or more time points via one or more sensors installed in an in-use vehicle; An input having a processor having at least one input and at least one output receiving the locations of the object detected by the one or more sensors; detecting the detected locations Estimating a previous movement trajectory of the object relative to the vehicle; using the estimated previous movement trajectory to predict the future position of the object relative to the vehicle; estimating the vehicle based on the predicted future position of the object relative to the vehicle a collision probability of the object; determining whether the collision probability exceeds a predetermined threshold; and outputting an early warning signal from an output of the processor if the collision probability exceeds the predetermined threshold. The method of collision avoidance of claim 1, wherein the one or more sensors are mounted along a horizontal axis of the vehicle side of the vehicle in use. The collision avoidance method of claim 1 or 2, wherein the one or more sensors comprise at least two distance sensors. The collision avoidance method of claim 1 or 2, wherein the one or more sensors comprise at least one position sensor. The anti-collision method of claim 1 or 2, wherein the detecting step is performed at three or more time points. The anti-collision method of claim 1 or 2, wherein the method further comprises filtering out irrelevant detection to obtain a series of useful detection steps. The anti-collision method of claim 1 or 2, wherein the method further comprises: differentiating the sensor detection according to the sensor ID and the detection in the case that there are multiple objects in the scene; a subset of. The anti-collision method of claim 1 or 2, wherein the method further comprises selecting a set of cameras for photographing a scene adjacent to the vehicle, so that the image processed by the algorithm can distinguish various types of objects. And the resulting information is used to assist in the sensor ID grouping procedure. The collision avoidance method of claim 1 or 2, wherein the method further comprises assigning a specific ID to each object and tracking the object ID for subsequent processing. The anti-collision method of claim 1 or 2, wherein the step of estimating the previous movement trajectory of the object is based on an optimization method, the optimization method is to compare the object with respect to the vehicle One of the cost functions of the previous movement trajectory is minimized to select a set of azimuth angles that are most likely to be associated with the one or more sensors at each of the points in time. The anti-collision system of claim 10, wherein the optimization method is a quadratic programming method. The collision avoidance system of claim 10, wherein the optimizing method comprises analyzing the series of detections to establish inequality constraints. The collision avoidance system of claim 10, wherein when the one or more sensors comprise at least two distance sensors, the optimizing method comprises measuring with two adjacent sensors The resulting distance is triangulated for the position of the object to set equality constraints. The collision avoidance method of claim 1 or 2, wherein the method further comprises the step of smoothing the estimated previous trajectory of the object using a Kalman filter. The collision avoidance method of claim 1 or 2, wherein the step of predicting a future movement trajectory of the object is performed on the assumption that the acceleration and/or yaw rate of the object is constant with respect to the vehicle. The anti-collision method of claim 1 or 2, wherein the vehicle is a heavy goods vehicle. The anti-collision method according to claim 1 or 2, wherein the object to be detected is a two-wheeled motor vehicle or a pedestrian. The anti-collision method of claim 1 or 2, wherein the warning signal comprises a video and/or audio warning signal, and if the collision possibility exceeds the predetermined threshold, the method further comprises transmitting the warning signal Go to a video/audio alarm to activate the alarm. The anti-collision method of claim 1 or 2, wherein the warning signal includes a brake signal, the method further comprising: if the collision possibility exceeds the predetermined threshold, transmitting the brake signal to one of the vehicles The braking system is configured to initiate the braking of the vehicle; and/or the warning signal includes a steering signal, the method further comprising: if the collision probability exceeds the predetermined threshold, transmitting the steering signal to a control system of the vehicle, The step of changing the steering angle of the vehicle. A computer program product suitable for use in a collision avoidance method, the computer program product comprising: a storage device comprising a plurality of instructions that, when executed by the processor, enable the processor to execute any of the foregoing patent applications The method steps described in the scope. A vehicle collision avoidance system, the system comprising: one or more sensors mounted in an in-use vehicle for detecting the position of an object relative to the vehicle at two or more points in time; a processor having at least one input and at least one output, the processor configured to: receive, by the input of the processor, the locations detected by the one or more sensors Estimating the previous movement trajectory of the object relative to the vehicle based on the detected positions; using the estimated previous movement trajectory to predict the future position of the object relative to the vehicle; according to the predicted object relative to a future location of the vehicle, estimating a likelihood of collision of the vehicle with the object; determining whether the likelihood of the collision exceeds a predetermined threshold; and outputting an early warning signal from an output of the processor if the likelihood of the collision exceeds the predetermined threshold . The collision avoidance system of claim 21, wherein the one or more sensors are mounted for use along a horizontal axis of the side of the vehicle in use. The collision avoidance system of claim 21 or 22, wherein the one or more sensors are respectively used at a height of 0.5 m to 1.5 m from the ground. The collision avoidance system of claim 21 or 22, wherein the one or more sensors comprise at least two distance sensors. The collision avoidance system of claim 21 or 22, wherein the one or more sensors comprise at least one position sensor. The collision avoidance system of claim 21, wherein the system comprises two or more sensors and the distance between two adjacent sensors is about 0.8 meters. The anti-collision system of claim 21 or 22, wherein the processor is configured to perform the detecting step at three or more points in time. The anti-collision system of claim 21, wherein the processor is further configured to perform an unrelated detection to obtain a series of useful detection steps. The collision avoidance system of claim 21 or 22, wherein the processor is configured to estimate based on an optimization method that minimizes a cost function of the object relative to the previous movement trajectory of the vehicle The previous movement trajectory of the object to select a set of azimuth angles most likely to be associated with the one or more sensors at various points in time. The anti-collision system of claim 29, wherein the optimization method is a quadratic programming method. The collision avoidance system of claim 29, wherein the optimizing method comprises analyzing the series of detections to establish inequality constraints. The collision avoidance system of claim 29, wherein when the one or more sensors comprise at least two distance sensors, the optimizing method comprises measuring with two adjacent sensors. The distance is triangulated for the position of the object to set the equality constraints. The collision avoidance system of claim 21 or 22, wherein the processor is configured to smooth the estimated previous movement trajectory of the object using a Kalman filter. The anti-collision system of claim 29, wherein the processor predicts a future movement trajectory of the object under the assumption that the acceleration and/or yaw rate of the object is constant with respect to the vehicle. The collision avoidance system of claim 21 or 22, wherein the vehicle is a heavy goods vehicle. The anti-collision system of claim 21 or 22, wherein the object to be detected is a two-wheeled motor vehicle or a pedestrian. The collision avoidance system of claim 21 or 22, wherein the system further comprises a guide cone for one of the at least one or each of the sensors. The anti-collision system of claim 21 or 22, wherein the warning signal comprises a video and/or audio warning signal, the system further comprising an alarm, if the collision possibility exceeds the predetermined threshold, the The processor further transmits the warning signal to the alarm to activate the alarm. The anti-collision system of claim 21 or 22, wherein the warning signal comprises a brake signal, and if the collision possibility exceeds the predetermined threshold, the processor further transmits the warning signal to one of the vehicle brakes The system is to hold the vehicle; and/or the warning signal includes a turn signal, and if the collision probability exceeds the predetermined threshold, the processor further transmits the warning signal to one of the vehicle control systems to change the steering of the vehicle angle.