Unmanned vehicle obstacle avoidance algorithm design based on multi-radar fusion

Abstract

According to the invention, the unmanned vehicle obstacle avoidance algorithm design based on multi-radar fusion is realized. According to the algorithm, a single-line radar is used for detecting front obstacles, and parking or reversing is judged through a minimum distance threshold value; obstacle breakpoints and angular points are found in the left part and the right part of the rectangular area, obtained point cloud data are filtered and reordered, point cloud is segmented through the nearest neighbor and IEPF fusion algorithm to obtain a line segment set, and the breakpoint angles and distances are recorded; calculating the distance of a breakpoint connecting line in the x direction, and judging whether the distance is greater than a passable threshold value or not; if there are passable areas on the left and right sides, selecting an obstacle avoidance direction according to the radar average distance; obstacle bypassing to a selected area, and maintaining a path by using a side ultrasonic sensor; and after bypassing the obstacle, returning to the original route when the radar cannot detect the obstacle. A real vehicle test shows that the obstacle avoidance of the unmanned vehicle under the static obstacle can be realized by utilizing the fusion perception of the single-line laser radar and the ultrasonic wave.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of autonomous vehicles, and particularly to a design of an obstacle avoidance algorithm for autonomous vehicles based on multi-radar fusion. Background Art

[0002] In the current era of rapid technological development, the technology of autonomous vehicles in the field of intelligent transportation has become a hot topic of research and application. With the continuous improvement of the logistics industry's demand for distribution efficiency and cost control, as well as the urgent requirements for intelligent and unmanned operations in scenarios such as urban sanitation and park patrol, autonomous vehicles have been widely used in many fields such as logistics distribution, intelligent sanitation, and park patrol due to their automation and high efficiency. Taking logistics distribution as an example, autonomous vehicles can accurately deliver goods within areas such as industrial parks and communities according to preset routes, greatly reducing labor costs and improving distribution efficiency; in the intelligent sanitation scenario, autonomous cleaning vehicles can independently plan cleaning routes and perform cleaning operations on areas such as urban streets and parks, realizing the intelligence and unmanned operation of sanitation work.

[0003] However, in the actual driving process of autonomous vehicles, obstacle avoidance technology is the core key to ensuring their safe and stable operation. At present, most common obstacle avoidance solutions for autonomous vehicles rely on a single sensor. Among them, single-line lidar has become one of the more commonly used sensors due to its high accuracy in ranging and relatively long detection distance. It emits laser beams and receives reflected signals to construct a point cloud model of the surrounding environment, thereby realizing the detection of obstacles. However, in complex environments, such as scenes with strong light changes and a large number of objects with similar reflectivities, the obstacle recognition ability of single-line lidar will be significantly affected, and false positives or false negatives are likely to occur. For example, on a road under direct sunlight, the reflected signals of the lidar may be interfered, resulting in misidentifying some non-obstacle objects as obstacles or missing some small obstacles.

[0004] Ultrasonic sensors are also widely used in obstacle avoidance for autonomous vehicles because of their simple structure, low cost, and certain real-time performance in short-range detection. It uses the reflection principle of ultrasonic waves to measure the distance between the sensor and the obstacle. However, ultrasonic sensors have obvious limitations. Their detection range is limited, generally with an effective detection distance within a few meters, and the accuracy will drop significantly in long-range detection. When an autonomous vehicle is driving on an open road and needs to detect obstacles at a relatively long distance, ultrasonic sensors often cannot provide accurate and reliable data.

[0005] In addition to the limitations of single sensors, existing obstacle avoidance algorithms also have deficiencies in path planning and obstacle avoidance decision-making. Most traditional obstacle avoidance algorithms make decisions based on simple threshold judgments or fixed rules, lacking in-depth analysis and comprehensive utilization of environmental information. For example, when faced with multiple obstacles, some algorithms are unable to quickly and accurately plan the optimal obstacle avoidance path, resulting in a long driving route and low efficiency during the obstacle avoidance process of the unmanned vehicle; some other algorithms do not fully consider the dynamic characteristics and driving state of the unmanned vehicle itself during obstacle avoidance decision-making, which may lead to sharp steering or acceleration / deceleration, not only affecting the riding comfort but also increasing the risk of collision with obstacles.

[0006] In summary, there are certain defects in the current obstacle avoidance technology of unmanned vehicles in terms of sensor application and algorithm design, which cannot meet the requirements of increasingly complex actual application scenarios. Therefore, there is an urgent need for an algorithm that can fuse multiple sensor information to achieve efficient and accurate obstacle avoidance, so as to improve the safety and reliability of unmanned vehicles in various environments. Summary of the Invention

[0007] The purpose of the present invention is: to overcome the above problems, provide a design of an obstacle avoidance algorithm for unmanned vehicles based on multi-radar fusion, and achieve a stable and accurate obstacle avoidance task for unmanned vehicles.

[0008] To achieve the above purpose, the present invention provides a design of an obstacle avoidance algorithm for unmanned vehicles based on multi-radar fusion, including the following steps:

[0009] Step 1: Use a single-line radar to detect obstacles in front. If the distance is less than the set threshold, stop the vehicle. If the stopping distance is less than the minimum obstacle avoidance distance, reverse until a safe distance is reached.

[0010] Step 2: Filter and reorder the obtained point cloud data.

[0011] Step 3: Extract breakpoints in the point cloud through the nearest neighbor algorithm, and divide the point cloud into several line sets according to the breakpoints.

[0012] Step 4: Use the IEPF algorithm to extract corner points in each line, and divide the point cloud into multiple line segment sets through the combination of corner points and breakpoints.

[0013] Step 5: Search for obstacle breakpoints in the left and right parts of the rectangular area, and record the breakpoint angles and distances.

[0014] Step 6: If there are passable areas on both the left and right, select the obstacle avoidance direction according to the average radar distance.

[0015] Step 7: Avoid obstacles towards the selected area, and use the side ultrasonic sensors to maintain the path.

[0016] Step 8: After bypassing the obstacle, when the radar cannot detect the obstacle, return to the original route.

[0017] Optionally, step 1 specifically includes:

[0018] First, the vehicle uses a single-line radar to detect obstacles in the front rectangular area in real time. When the detected distance is less than the set threshold, it stops and switches to the obstacle avoidance mode. Specifically, the single-line radar receives data through the serial port, and the stm32 parses the original data. Downsampling is achieved by skip point sampling, and the angle and distance corresponding to each point are synthesized. When the stm32 receives all the data of one circle of the radar, the point cloud is grouped, and the closest one of the ten points is retained as the final selected point of each group. Then, the distances mindis[i] and angles minang[i] of each group of selected points are traversed to find the coordinates of the points falling within the rectangular area. For the coordinates within the rectangular area, the minimum distance in the y direction is used as the obstacle detection distance and compared with the parking threshold.

[0019] Note: The area can be selected as a rectangle (|x| ≤ x set , y ≥ y set ), a 90° front fan (-45° < mindis[i] < 45°), and a 150° front fan area (-72.5° < mindis[i] < 72.5°)

[0020] The coordinate calculation formula within the rectangular area is as follows:

[0021]

[0022] Record the minimum distance of obstacle detection in the front rectangular area after parking. If it does not meet the minimum distance for obstacle avoidance, reverse until it stops when greater than the threshold. Specifically, we compare the minimum distance y min obtained in step 1 with the minimum obstacle avoidance threshold. If y min is less than the minimum obstacle avoidance threshold, the vehicle performs the reverse task until y min detected by the front radar is greater than the obstacle avoidance threshold and then stops.

[0023] Optionally, step 2 specifically includes:

[0024] In the preprocessing stage of point cloud data, first, it is necessary to filter the single-line lidar point cloud data received from the serial port. The main purpose is to filter out the data with a point distance of 0, because this data is usually invalid or noise points caused by sensor errors. After completing the preliminary filtering, starting from the first valid point in the point cloud data, calculate the distance between two adjacent points one by one. When it is found that the distance is greater than the preset threshold for the first time, mark the next point as the new starting point. This process helps to identify the breakpoints or segmentation points in the point cloud data, providing a reference for subsequent processing. Subsequently, sort the point cloud data according to the scanning order of the lidar and add an index to each point for subsequent analysis and processing.

[0025] Optionally, step 3 specifically includes:

[0026] When the vehicle is in a position where it can avoid obstacles, first divide the rectangular area in front into left and right parts to process the environmental information on both sides separately. Next, search for breakpoints in each part of the rectangular area, record the angles and distances of these breakpoints, and perform index matching. The specific operation is to use the nearest neighbor algorithm to find breakpoints in the point cloud data. This algorithm determines whether there is a breakpoint by calculating the Euclidean distance between adjacent points in the point cloud. If the distance between two points is greater than the preset threshold, mark this point as a breakpoint. According to these breakpoints, divide the point cloud data into several line sets composed of points. Each line has a clear head and tail, so breakpoints appear in pairs, divided into head breakpoints and tail breakpoints. In the point cloud data, the starting point is defaulted as the head breakpoint of the first line, and the ending point is the tail breakpoint of the last line. Finally, re-index and mark all breakpoints from -90° to 90° in the radar scanning direction.

[0027] Optionally, step 4 specifically includes:

[0028] In the divided set of lines, corner points of each line are extracted through the IEPF algorithm. First, the starting point and the ending point of each line segment are connected to form a straight line. This straight line can serve as the initial reference line. Then, the perpendicular distance from all points on the line segment to this straight line is calculated. This can be achieved by calculating the shortest distance from each point to the straight line. The point with the maximum distance is found and regarded as a corner point of the line segment. This point usually represents the position where the direction of the line segment changes significantly. Taking this corner point as the demarcation point, the original line segment is divided into two new line segments. Each new line segment contains the part from the starting point to the corner point and the part from the corner point to the ending point. The above steps are repeated for these two new line segments, continuing to find corner points and divide the line segments. This process is recursive and stops iterating until the distance from the found corner point to the straight line is less than a preset threshold. This threshold can be set according to the requirements of specific applications to ensure that the extracted corner points have sufficient accuracy and representativeness. Finally, all the extracted corner points are combined with the original starting points and ending points to form multiple sets of line segments composed of points. These sets of line segments can be sorted according to the order of lidar scanning to better reflect the structure and characteristics of the point cloud data.

[0029] Optionally, step 5 specifically includes:

[0030] Connect the breakpoints in pairs and calculate the distance in the x direction to determine whether there is a situation where it is greater than the passable threshold. The calculation formula is as follows:

[0031] diff_x = g_pointx[i] - g_pointx[i - 1]

[0032] At this time, when diff_x is greater than the threshold of the set vehicle width plus the inflation radius, it represents a passable area.

[0033] Optionally, step 6 specifically includes:

[0034] If there is a passable area on either the left or the right, the vehicle bypasses the obstacle towards the passable area. If there are passable areas in both the left front area and the right front area, the bypass direction is determined by judging the average distance of a circle of radar in the two areas; specifically, in this paper, the rectangular area is divided into the left front area and the right front area. When there is only one passable area, it is marked as the only passable area; when there are passable areas in both the left front and the right front, in this paper, the point cloud distances in the two areas are averaged respectively, and the area with the larger average value is selected as the only passable area.

[0035] Optionally, step 7 specifically includes:

[0036] The vehicle detects obstacles through the front radar to determine whether the initial obstacle avoidance is completed, and uses the ultrasonic sensors on the sides to maintain obstacle avoidance. Specifically, when the obstacle detection distance on one side of the obstacle by the front radar is greater than the parking threshold, it is determined that the initial obstacle avoidance is completed; when the initial obstacle avoidance is completed, the vehicle starts to turn the steering wheel towards the side of the obstacle, and when the vehicle's heading angle is consistent with the original route, the straight-line task starts. At this time, the ultrasonic sensors continuously monitor the distance to the obstacle. When the average distance is less than the safe distance, the vehicle makes fine adjustments through the PID algorithm;

[0037] The following is the calculation formula of the PID algorithm:

[0038]

[0039] where e(t) is the error signal, and e(t) = θ desired - θ actual , θ desired is the desired angle, and θ actual is the actual angle.

[0040] Optionally, step 8 specifically includes:

[0041] When the front radar cannot detect the current obstacle, the vehicle starts to return to the original route. Specifically, when the obstacle detection distance on one side of the obstacle by the front radar is greater than the threshold, the vehicle drives in the opposite direction until its heading angle is the same as that of the original route, and the obstacle avoidance task is completed.

[0042] The present invention realizes a design of an obstacle avoidance algorithm for an autonomous vehicle based on multi-radar fusion. The algorithm uses a single-line radar to detect obstacles in front. If the distance is less than the set threshold, the vehicle stops; if the stopping distance is less than the minimum obstacle avoidance distance, it reverses until the safe distance. It searches for obstacle breakpoints and corner points in the left and right parts of the rectangular area, filters and reorders the obtained point cloud data, segments the point cloud through the nearest neighbor and IEPF fusion algorithms to obtain a set of line segments, and records the breakpoint angles and distances; calculates the distance of the breakpoint connection in the x direction to determine whether there is a situation greater than the passable threshold; if there are passable areas on both the left and right, it selects the obstacle avoidance direction according to the average radar distance; avoids obstacles towards the selected area and uses the side ultrasonic sensors to maintain the path; after bypassing the obstacle, when the radar cannot detect the obstacle, it returns to the original route.

[0043] The actual vehicle tests show that the present invention can use the fusion perception of a single-line lidar and ultrasonic sensors to achieve obstacle avoidance of an autonomous vehicle under static obstacles. Description of the Drawings

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following briefly introduces the accompanying drawings required for use in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0045] Figure 1 is the flow schematic diagram of the present invention.

[0046] Figure 2 is the schematic diagram of obstacle recognition under a rectangular area.

[0047] Figure 3 is the schematic diagram of the obstacle avoidance planning route.

[0048] Figure 4 is the schematic diagram of extracting the break points of the point cloud.

[0049] Figure 5 is the schematic diagram of extracting the corner points of the point cloud.

[0050] Figure 6 is the schematic diagram of the second extraction of corner points of the present invention. Detailed implementation manners

[0051] The following details the embodiments of the present invention. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention, and should not be construed as a limitation of the present invention.

[0052] Please refer to Figure 1 , the present invention provides an obstacle avoidance algorithm design for an autonomous vehicle based on multi-radar fusion, including the following steps:

[0053] S1: Use a single-line radar to detect obstacles ahead. If the distance is less than the set threshold, stop the vehicle. If the stopping distance is less than the minimum obstacle avoidance distance, reverse until a safe distance is reached.

[0054] S2: Filter and reorder the obtained point cloud data.

[0055] S3: Extract the break points in the point cloud through the nearest neighbor algorithm, and divide the point cloud into several line sets according to the break points.

[0056] S4: Extract the corner points in each line using the IEPF algorithm, and divide the point cloud into multiple line segment sets through the combination of corner points and break points.

[0057] S5: Search for obstacle break points in the left and right parts of the rectangular area, and record the break point angles and distances.

[0058] S6: If there are passable areas on both the left and right, select the obstacle avoidance direction according to the average radar distance.

[0059] S7: Avoid obstacles towards the selected area and maintain the path using the side ultrasonic sensors.

[0060] S8: After bypassing the obstacle, when the radar can no longer detect the obstacle, return to the original route.

[0061] The following is a further description in combination with specific implementation steps:

[0062] S1: First, the vehicle uses a single-line radar to detect obstacles in the front rectangular area in real time. When the detected distance is less than the set threshold, it stops and switches to the obstacle avoidance mode. Specifically, the single-line radar receives data through the serial port, and the stm32 parses the original data, performs downsampling by skip point sampling, and synthesizes the angle and distance corresponding to each point. When the stm32 finishes receiving the data of one circle of the radar, it groups the point cloud and retains the nearest point from every ten points as the final selected point of each group. Then, it traverses the distance mindis[i] and angle minang[i] of each selected point group to find the coordinates of the points falling within the rectangular area. For the coordinates within the rectangular area, the minimum distance in the y direction is used as the obstacle detection distance for comparison with the parking threshold.

[0063] Note: The area can be selected as a rectangle (|x| ≤ x set , y ≥ y set ), a 90° front sector (-45° < mindis[i] < 45°), and a 150° front sector area (-72.5° < mindis[i] < 72.5°)

[0064] The coordinate calculation formula within the rectangular area is as follows:

[0065]

[0066] Record the minimum distance of obstacle detection in the front rectangular area after parking. If it does not meet the minimum distance for obstacle avoidance, reverse the vehicle until it stops when the distance is greater than the threshold. Specifically, we compare the minimum distance y min obtained in step 1 with the minimum obstacle avoidance threshold. If y min is less than the minimum obstacle avoidance threshold, the vehicle performs the reverse task until the y detected by the front radar min is greater than the obstacle avoidance threshold and then stops.

[0067] S2: In the point cloud data preprocessing stage, first, it is necessary to filter the single-line lidar point cloud data received from the serial port. The main purpose is to filter out the data with a point distance of 0, because these data are usually invalid or noise points caused by sensor errors. After completing the preliminary filtering, starting from the first valid point in the point cloud data, calculate the distance between adjacent two points one by one. When it is found that the distance is greater than the preset threshold for the first time, mark the next point as the new starting point. This process helps to identify the breakpoints or segmentation points in the point cloud data, providing a reference for subsequent processing. Subsequently, sort the point cloud data according to the lidar scanning order and add an index to each point for subsequent analysis and processing.

[0068] S3: When the vehicle is in a position where it can bypass obstacles, first divide the rectangular area in front into left and right parts to process the environmental information on both sides separately. Next, search for breakpoints in each part of the rectangular area, record the angles and distances of these breakpoints, and perform index matching. The specific operation is to use the nearest neighbor algorithm to find breakpoints in the point cloud data. This algorithm determines whether there is a breakpoint by calculating the Euclidean distance between adjacent two points in the point cloud. If the distance between two points is greater than the preset threshold, mark this point as a breakpoint. According to these breakpoints, divide the point cloud data into several line sets composed of points. Each line has a clear head and tail, so breakpoints appear in pairs, divided into head breakpoints and tail breakpoints. In the point cloud data, the starting point is defaulted as the head breakpoint of the first line, and the ending point is the tail breakpoint of the last line. Finally, re-index and mark all breakpoints from -90° to 90° in the radar scanning direction.

[0069] S4: In the divided line sets, extract the corner points of each line through the IEPF algorithm. First, connect the starting point and the ending point of each line segment to form a straight line. This straight line can be used as the initial reference line. Then, calculate the perpendicular distance from all points on this line segment to this straight line. This can be achieved by calculating the shortest distance from each point to the straight line. Find the point with the maximum distance and regard it as a corner point of this line segment. This point usually represents the position where the direction of the line segment changes significantly. Taking this corner point as the demarcation point, divide the original line segment into two new line segments. Each new line segment contains the part from the starting point to the corner point and the part from the corner point to the ending point. Repeat the above steps for these two new line segments, continue to find corner points and divide the line segments. This process is recursive and stops iterating until the distance from the found corner point to the straight line is less than a preset threshold. This threshold can be set according to the specific application requirements to ensure that the extracted corner points have sufficient accuracy and representativeness. Finally, combine all the extracted corner points with the original starting points and ending points to form multiple line sets composed of points. These line sets can be sorted according to the lidar scanning order to better reflect the structure and characteristics of the point cloud data.

[0070] S5: Connect the breakpoints in pairs and calculate the distance in the x - direction to determine if there is a case greater than the passable threshold. The calculation formula is as follows:

[0071] diff_x = g_pointx[i] - g_pointx[i - 1]

[0072] At this time, when diff_x is greater than the threshold of the set vehicle width plus the expansion radius, it represents a passable area.

[0073] S6: If there is a passable area on either the left or the right, the vehicle bypasses the obstacle towards the passable area. If there are passable areas in both the left - front area and the right - front area, the bypass direction is determined by the average distance of the radar around the two areas; specifically, in this paper, the rectangular area is divided into the left - front area and the right - front area. When there is only one passable area, it is marked as the only passable area; when there are passable areas in both the left - front and the right - front, the average value of the point - cloud distances of the two areas is calculated respectively in this paper, and the area with the larger average value is selected as the only passable area.

[0074] S7: The vehicle detects obstacles through the front radar to determine if the initial bypass is completed, and uses the side ultrasonic detection for bypass maintenance. Specifically, when the obstacle detection distance on one side of the obstacle by the front radar is greater than the parking threshold, it is judged that the initial bypass is completed; when the initial bypass is completed, the vehicle starts to turn the steering wheel towards the obstacle side. When the vehicle's heading angle is consistent with the original route, it starts the straight - line task. At this time, the ultrasonic wave continuously monitors the distance to the obstacle. When its average distance is less than the safety distance, the vehicle makes fine - tuning through the PID algorithm;

[0075] The following is the calculation formula of the PID algorithm:

[0076]

[0077] Among them, e(t) is the error signal, e(t)=θ desired - θ actual , θ desired is the desired angle, θ actual is the actual angle.

[0078] S8: When the front radar cannot detect the current obstacle, it starts to return to the original route. Specifically, when the obstacle detection distance on one side of the obstacle by the front radar is greater than the threshold, the vehicle drives in the opposite direction until its heading angle is the same as the original route heading angle, and the obstacle - bypass task is completed.

[0079] The above-disclosed is only a preferred embodiment of the present invention. Of course, the scope of rights of the present invention cannot be limited thereby. Those of ordinary skill in the art can understand all or part of the processes of implementing the above embodiments, and the equivalent changes made according to the claims of the present invention still fall within the scope covered by the invention.

Claims

1. Design of an obstacle avoidance algorithm for an unmanned vehicle based on multi-radar fusion, characterized in that It includes the following steps: Step 1: Use a single-line radar to detect obstacles ahead. If the distance is less than the set threshold, stop the vehicle. If the stopping distance is less than the minimum obstacle avoidance distance, reverse until a safe distance is reached. Step 2: Filter and reorder the obtained point cloud data. Step 3: Extract breakpoints in the point cloud through the nearest neighbor algorithm, and divide the point cloud into several line sets according to the breakpoints. Step 4: Use the IEPF algorithm to extract corner points in each line, and divide the point cloud into multiple line segment sets by combining corner points and breakpoints. Step 5: Search for obstacle breakpoints in the left and right parts of the rectangular area, and record the breakpoint angles and distances. Step 6: If there are passable areas on both the left and right, select the obstacle avoidance direction according to the average radar distance. Step 7: Avoid obstacles towards the selected area and maintain the path using side ultrasonic sensors. Step 8: After bypassing the obstacle, when the radar can no longer detect the obstacle, return to the original route.

2. For an obstacle avoidance algorithm design of an autonomous vehicle based on multi-radar fusion as described in claim 1, for the said step 1, it is characterized in that: First, the vehicle uses a single-line radar to continuously detect obstacles in the front rectangular area. When the detected distance is less than the set threshold, it stops and switches to the obstacle avoidance mode; specifically, the single-line radar receives data through the serial port, and the stm32 parses the original data, and implements downsampling through skip point sampling, synthesizing the angle and distance corresponding to each point; when the stm32 finishes receiving a full circle of radar data, it groups the point cloud, and retains the nearest point from every ten points as the final selected point of each group; then traverse the distance mindis[i] and angle minang[i] of each group of selected points to find the coordinates of the points falling within the rectangular area; for the coordinates within the rectangular area, use the minimum distance in the y direction as the obstacle detection distance and compare it with the stopping threshold. Note: The area can be selected as a rectangle (|x|≤x set , y≥y set ), a 90° front sector (-45° < mindis[i] < 45°), and a 150° front sector area (-72.5° < mindis[i] < 72.5°) The coordinate calculation formula within the rectangular area is as follows: Record the minimum distance for obstacle detection in the front rectangular area after parking. If it does not meet the minimum distance for obstacle avoidance, reverse the vehicle until it stops when the distance is greater than the threshold; specifically, we use the minimum distance y obtained in step 1 min to compare with the minimum threshold for obstacle avoidance. If y min is less than the minimum threshold for obstacle avoidance, the vehicle performs the reverse task until the y detected by the front radar min is greater than the obstacle avoidance threshold and then stops.

3. For an obstacle avoidance algorithm design of an autonomous vehicle based on multi-radar fusion as described in claim 1, for the said step 2, it is characterized in that In the preprocessing stage of the point cloud data, first filter the point cloud data of the single-line lidar received through the serial port, mainly filtering out the data with a point distance of 0, and then calculate the distance between adjacent two points starting from the first point of the point cloud data. Take the next point after the first distance greater than the threshold as the starting point, and sort the point cloud data according to the scanning sequence of the lidar and add indexes.

4. For an obstacle avoidance algorithm design of an autonomous vehicle based on multi-radar fusion as described in claim 1, for the said step 3, it is characterized in that When the vehicle is in the obstacle - bypassable position, divide the rectangular area into left and right parts, search for the breakpoints in the front rectangular area, record the angles and distances of these breakpoints, and perform index matching. Specifically, find the breakpoints in the point - cloud data through the nearest - neighbor algorithm. Mainly calculate whether the Euclidean distance between two adjacent points in the point cloud is greater than a certain threshold, and regard the points greater than the threshold as breakpoints. Divide the point cloud into several line sets composed of points. A line has a head and a tail, so breakpoints appear in pairs, divided into head breakpoints and tail breakpoints. The starting point in the point cloud is defaulted as the head breakpoint of the first line, and the ending point is used as the tail breakpoint of the last line. Re - index and mark the breakpoints from - 90° to 90° in the radar scanning direction.

5. The design of an obstacle - avoidance algorithm for an autonomous vehicle based on multi - radar fusion according to claim 1, wherein step 4 is characterized in that In the divided line sets, extract the corner points of each line through the IEPF algorithm. First, connect the starting point and the ending point of each line segment to form a straight line. This straight line can be used as the initial reference line. Then, calculate the perpendicular distance from all points on this line segment to this straight line. This can be achieved by calculating the shortest distance from each point to the straight line. Find the point with the largest distance and regard it as a corner point of this line segment. This point usually represents the position where the direction of the line segment changes significantly. Taking this corner point as the demarcation point, divide the original line segment into two new line segments. Each new line segment contains the part from the starting point to the corner point and the part from the corner point to the ending point. Repeat the above steps for these two new line segments, continue to find corner points and divide the line segments. This process is recursive and stops iterating until the distance from the found corner point to the straight line is less than a preset threshold. This threshold can be set according to the requirements of specific applications to ensure that the extracted corner points have sufficient accuracy and representativeness. Finally, combine all the extracted corner points with the original starting points and ending points to form multiple line - segment sets composed of points. These line - segment sets can be sorted according to the order of lidar scanning to better reflect the structure and characteristics of the point - cloud data.

6. The design of an obstacle - avoidance algorithm for an autonomous vehicle based on multi - radar fusion according to claim 1, wherein step 5 is characterized in that Connect the breakpoints in pairs and calculate the distance in the x - direction to determine whether there is a situation where it is greater than the passable threshold. The calculation formula is as follows: diff_x = g_pointx[i] - g_pointx[i - 1] At this time, when diff_x is greater than the threshold of the vehicle width plus the inflation radius, it represents a passable area.

7. The design of an obstacle - avoidance algorithm for an autonomous vehicle based on multi - radar fusion according to claim 1, wherein step 6 is characterized in that If there is a passable area on either the left or the right, the vehicle will bypass the obstacle towards the passable area. If there are passable areas in both the left front area and the right front area, the bypass direction will be determined by judging the average distance of the radar around the two areas; specifically, in this article, the rectangular area is divided into the left front area and the right front area. When there is only one passable area, it is marked as the only passable area; when there are passable areas in both the left front and the right front, in this article, the point cloud distances of the two areas are averaged respectively, and the area with the larger average value is selected as the only passable area.

8. For a multi-radar fusion-based obstacle avoidance algorithm design for driverless vehicles as described in claim 1, step 7 is characterized in that The vehicle detects obstacles through the front radar to determine whether the initial bypass is completed, and maintains the bypass through the ultrasonic detection on the side. Specifically, when the obstacle detection distance of the front radar on one side of the obstacle is greater than the parking threshold, it is judged that the initial bypass is completed; when the initial bypass is completed, the vehicle starts to turn the steering wheel towards the side of the obstacle. When the vehicle's heading angle is the same as the original route, the straight-line task starts. At this time, the ultrasonic wave constantly monitors the distance of the obstacle. When its average distance is less than the safe distance, the vehicle makes fine-tuning through the PID algorithm; The following is the calculation formula of the PID algorithm: where, e(t) is the error signal, and e(t) = θ desired - θ actual , θ desired is the desired angle, and θ actual is the actual angle.

9. For a multi-radar fusion-based obstacle avoidance algorithm design for driverless vehicles as described in claim 1, step 8 is characterized in that When the front radar cannot detect the current obstacle, it starts to bypass back to the original route. Specifically, when the obstacle detection distance of the front radar on one side of the obstacle is greater than the threshold, the vehicle drives in the opposite direction until its heading angle is the same as the heading angle of the original route, and the obstacle bypass task is completed.

Images