Event Camera-Based Human-Free Obstacle Avoidance Method, Device, and Intelligent Human-Free Body

Abstract

The present invention discloses a method, device and intelligent unmanned body for obstacle avoidance without human body based on an event camera. The method includes: using a binocular event camera to collect event images of moving obstacles in real time; calculating and saving the center point coordinates of the moving obstacles in the left event camera image and the right event camera image at each moment, as well as the size information and depth information of the moving obstacles at each moment; reading the center point coordinates of the moving obstacles in the continuous frame event images of any camera, as well as the size information and depth information of the moving obstacles, and calculating the motion vector of the target obstacle; and setting an obstacle avoidance trigger condition for the unmanned body according to the size information, motion vector of the target obstacle and the position of the target obstacle in the event image, and calculating the obstacle avoidance direction and motion distance. The present invention can achieve obstacle avoidance for fast-moving obstacles, effectively avoid unnecessary excessive obstacle avoidance maneuver behaviors, and has a small calculation cost.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer vision, and particularly relates to an obstacle avoidance method and device based on an event camera, as well as an intelligent unmanned body. Background Art

[0002] Intelligent unmanned bodies such as unmanned aerial vehicles and unmanned vehicles have the characteristics of being portable, flexible, highly maneuverable, and having good concealment, and are widely used in civilian and military fields. With the continuous development of unmanned bodies, the complex and changeable task environment makes the preset global trajectory unable to fully ensure the safety of unmanned bodies during the autonomous task execution process. Therefore, the autonomous obstacle avoidance technology of unmanned bodies in the face of sudden obstacles is a key component of the unmanned body system.

[0003] Currently, there are many target detection solutions based on traditional vision cameras in computer vision tasks. However, traditional vision cameras will produce motion blur for high-speed moving objects, cannot significantly present target objects in scenes with low light brightness, and cannot accurately detect targets in scenes where the target object is similar to the background, making it difficult for target detection methods based on traditional camera images to timely and accurately detect obstacles in complex environments and scenes with sudden obstacles.

[0004] Compared with traditional vision cameras, event cameras have the characteristics of low latency, high dynamic range, no motion blur, and ultra-low power consumption, and are commonly used in tasks with low light, high dynamics, or capturing high-speed moving objects. For example, Chinese patent document CN112200856A discloses a visual ranging method based on an event camera, which uses the tiny yolov3 algorithm for target detection to obtain the category of the target and its position on the image, and calculates the distance between the target and the camera using the similar triangle algorithm.

[0005] However, although this method can provide relatively high ranging accuracy, it still requires prior information of the target, such as parameters such as height and width. At the same time, this method only obtains the target distance and cannot solve the problem of how to avoid obstacles for moving obstacles. Summary of the Invention

[0006] The present invention aims to disclose an intelligent unmanned body obstacle avoidance method and device based on an event camera, which are used to solve the problems of measuring the size and distance of moving obstacles with unknown sizes, and avoiding obstacles for high-speed moving obstacle targets.

[0007] According to the first aspect of the present invention, an obstacle avoidance method for an unmanned body is disclosed. The unmanned body includes a binocular event camera, and the method includes:

[0008] Using the binocular event camera to collect event images of moving obstacles in real time;

[0009] Calculate and save the center point coordinates of moving obstacles in the left event camera image and the right camera event image at each moment, as well as the size information and depth information of the moving obstacles at each moment;

[0010] Read the center point coordinates of the moving obstacles in the continuous frame event images of any camera, as well as the size information and depth information of the moving obstacles, and calculate the motion vector of the target obstacle; and

[0011] Set the trigger condition for obstacle avoidance without a human body according to the size information, motion vector of the target obstacle, and the position of the target obstacle in the event image, and calculate the obstacle avoidance direction and the motion distance.

[0012] In some other examples, use the binocular event camera to collect the event images of the moving obstacles in real time, and input the obtained left event image and right event image at each moment into the neural network respectively to obtain the vertex coordinates of the target box of the moving obstacle in its respective event image.

[0013] In some other examples, calculate the center point coordinates of the moving obstacle in its respective event image using the vertex coordinates of the target box, and calculate the length, width, and depth information of the moving obstacle based on the principle of similar triangles.

[0014] In some other examples, set two warning areas with different sizes centered on the center point of the event image, and set the trigger condition for obstacle avoidance according to the relative position between the center point of the moving obstacle and the two warning areas.

[0015] In some other examples, when the center point of the moving obstacle is located in the event image, but outside the large warning area and the depth is less than the first threshold, trigger obstacle avoidance without a human body.

[0016] In some other examples, when the center point of the moving obstacle is located within the large warning area, the depth change value is less than 0 and the depth is less than the second threshold, trigger obstacle avoidance without a human body when any of the following conditions is met: (i) the moving obstacle moves in the direction of no human body in any dimension direction of the imaging plane; (ii) the center point of the moving obstacle is located within the small warning area.

[0017] In some other examples, after the obstacle avoidance is triggered, use the orthogonal vector of the motion vector of the moving obstacle as the obstacle avoidance direction without a human body.

[0018] In some other examples, after the obstacle avoidance is triggered, the obstacle avoidance motion distance is the sum of the actual physical size of the no human body and the larger value of the length and width of the moving obstacle.

[0019] According to the second aspect of the present invention, there is also disclosed an obstacle avoidance device for a no human body, and the device includes:

[0020] A binocular event camera for real-time acquisition of event images of moving obstacles;

[0021] A first computing unit for calculating and storing the center point coordinates of moving obstacles in the left event camera image and the right camera event image at each moment, as well as the size information and depth information of the moving obstacles at each moment;

[0022] A second computing unit for reading the center point coordinates of the moving obstacles, as well as the size information and depth information of the moving obstacles, in the continuous frame event images of any camera, and calculating the motion vector of the target obstacle; and

[0023] An obstacle avoidance decision-making unit for setting a human-free obstacle avoidance trigger condition according to the size information, motion vector of the target obstacle, and the position of the target obstacle in the event image, and calculating the obstacle avoidance direction and motion distance.

[0024] According to the third aspect of the present invention, there is also disclosed an intelligent human-free body, including a body, a driving device for driving the movement of the body, and a control device. It further includes a binocular event camera for collecting information on moving obstacles. The control device includes a processor and a memory. A computer program is stored in the memory, and the processor is used to execute the computer program to implement the obstacle avoidance method according to any one of the above solutions.

[0025] Compared with the prior art, the present invention can complete the obstacle avoidance of fast-moving obstacles by virtue of the characteristics of high dynamic range, low latency, and high dynamic performance of the event camera. At the same time, while ensuring reliable obstacle avoidance, it can effectively avoid unnecessary excessive obstacle avoidance maneuvering behaviors. And it has the characteristics of low computational cost, can output obstacle avoidance decisions within dozens of microseconds, is suitable for lightweight platforms, and can achieve autonomous obstacle avoidance by the on-board computing resources of the human-free body. Description of the Drawings

[0026] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, purposes, and advantages of the present invention will become more obvious:

[0027] Figure 1 A schematic method flow for training a neural network;

[0028] Figure 2 A schematic method flow for calculating the size and depth information of moving obstacles using the parallax method;

[0029] Figure 3 A schematic flow of the human-free body obstacle avoidance method according to an embodiment of the present invention;

[0030] Figure 4 A schematic diagram of setting the human-free body obstacle avoidance trigger condition of the present invention;

[0031] Figure 5 Schematic diagram of the composition of the human-free obstacle avoidance device according to an embodiment of the present invention;

[0032] Figure 6 Schematic diagram of the composition of the intelligent human-free device according to an embodiment of the present invention;

[0033] Figure 7 An implementation example of implementing human-free obstacle avoidance using the present invention. Detailed implementation manners

[0034] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention, rather than limiting the invention. Additionally, it should be noted that for the convenience of description, only the parts related to the invention are shown in the drawings.

[0035] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and embodiments.

[0036] The intelligent human-free device in the present invention can be an unmanned device traveling on land, in the air or in water, such as a drone, an autonomous vehicle, a robot, an unmanned ship, an underwater unmanned vehicle, etc. These intelligent human-free devices have a driving device for driving the movement of the human-free device, a sensor for collecting environmental information, and a control device for controlling the movement of the human-free device. The control device generally includes a processor, a memory, etc.

[0037] In the present invention, the human-free device is provided with a binocular event camera, that is, it includes a left camera and a right camera. The centers of the two cameras are separated by a predetermined distance, that is, the baseline length B. Generally, the binocular event camera can be horizontally placed on the human-free device and face directly forward, and is horizontally calibrated and registered.

[0038] The event camera is a camera that is only sensitive to the change in pixel brightness and can provide a response signal at the microsecond level. The human-free device equipped with the binocular event camera can complete the ranging of sudden targets (moving obstacles) and the effective perception of the environment in complex scenarios such as low light and high dynamics, so as to complete the obstacle avoidance task only by using limited on-board computing resources, increase the success rate of the human-free device to autonomously execute complex tasks, and improve the traffic safety of the human-free device.

[0039] First, the present invention uses the binocular event camera to collect event data, generate event images, and construct a data set to train the object detection neural network. Specifically, as Figure 1 shown, it includes the following steps:

[0040] S11: Use the event camera to collect the event stream generated by the rapid movement of the moving obstacle;

[0041] Each collected event is represented by a quadruple (t, x, y, p). Here, t represents the time when the event occurs, (x, y) represents the horizontal and vertical coordinates of the event occurrence location, and p represents the polarity of the event. When the brightness increase exceeds the threshold, the polarity is 1; when the brightness decrease exceeds the threshold, the polarity is 0. When the brightness change does not exceed the threshold, no event is generated.

[0042] S12: Generate event images from the event stream at fixed time intervals;

[0043] Based on the event generation mechanism, using the fixed time interval method, for each fixed time interval all event streams within it are used to generate event images. Then, the nth event image contains all events within the time period. Among them, the following method is used to generate event images: According to the pixel positions where events are generated, the coordinates with polarity generation are plotted as white pixels, and the background color of the image is black.

[0044] S13: Mark the target positions and categories in the event images to construct a training dataset;

[0045] The target position can be represented by a four-element array, corresponding to the maximum abscissa, minimum abscissa, maximum ordinate, and minimum ordinate of the four vertices of the target box respectively. The marked event images are divided into three parts: the training set, the test set, and the validation set. Among them, the training set accounts for 60%, and the test set and the validation set each account for 20% of the entire dataset.

[0046] S14: Use the dataset to train the YOLOV5 neural network to obtain the weight file of the neural network with the best performance on the test set;

[0047] It can be understood that other neural networks such as yolov3, yolov4, Faster R-CNN, and SSD can also be used in the present invention for object detection.

[0048] After the above steps, a neural network that can detect the category and position of moving obstacles in the images collected by the event camera is obtained. Specifically, a binocular event camera is used to collect event images of moving obstacles in real time without a human body, the weight file is loaded, and the category and position detection results of the moving obstacles are obtained using the neural network. Among them, the detection result of the target position is also represented by a four-element array (x right , x left , y down , y up) are output in the form of, corresponding to the maximum abscissa, minimum abscissa, maximum ordinate, and minimum ordinate of the four vertices of the result target box respectively. Then, the center point coordinates (X, Y) of the moving obstacle in the event image can be expressed by the vertex coordinates of the target detection box as follows:

[0049]

[0050] In the present invention, since there is no human body and a binocular event camera is set, the parallax method can be used to calculate the size of the moving obstacle and the distance between the moving obstacle and the human - free body. Specifically, as Figure 2 shown, the following steps are included:

[0051] S21: Use the binocular event camera to collect the event images of the moving obstacle in real - time, and input the left - hand event image and the right - hand event image obtained at each moment into the neural network respectively to obtain the vertex coordinates of the target box of the moving obstacle in their respective event images at the same moment;

[0052] S22: Use the vertex coordinates of the target box to calculate the center point coordinates of the moving obstacle in their respective event images, and calculate the length and width dimensions of the moving obstacle and the distance between the moving obstacle and the human - free body based on the principle of similar triangles;

[0053] The parallax dis between the left - hand event image and the right - hand event image is;

[0054] dis = B-(X L -X R )

[0055] where B is the baseline length, that is, the distance between the center points of the left - hand and right - hand cameras, X L is the abscissa of the center point of the moving obstacle in the event image collected by the left - hand camera, and X R is the abscissa of the center point of the moving obstacle in the event image collected by the right - hand camera.

[0056] Based on the principle of similar triangles, it can be obtained that:

[0057]

[0058] Then:

[0059]

[0060] where f is the focal length of the event camera, Z is the depth information, that is, the distance between the moving obstacle and the center of the binocular camera. In the present invention, it is simplified to the distance between the moving obstacle and the human - free body. Those skilled in the art can easily understand that there is a fixed and simple conversion relationship between this calculated distance and the actual distance.

[0061] Based on the position detection result of the moving obstacle in the event image collected by either the left camera or the right camera, that is, the vertex coordinates of the target box, the length h and width w of the moving obstacle can be calculated based on the principle of similar triangles as follows:

[0062]

[0063]

[0064] where u is the size length of a pixel in the image.

[0065] Through the above method, by using a binocular camera, the present invention can obtain the center point coordinates, length and width size information of an unknown moving obstacle at each moment, as well as the distance information (depth information) of the target from the non-human body (the center of the binocular event camera), and at least save the center point coordinates, size information and depth information of the current k moments (that is, the nearest k ), 2 ≤ k ≤ n, preferably k = 2.

[0066] On this basis, the present invention proposes a human-free obstacle avoidance method for high-speed moving obstacles, as Figure 3 shown, the method includes the following steps:

[0067] S31: Calculate the motion vector of the target obstacle using consecutive frame event images;

[0068] Select two adjacent event images collected by either the left camera or the right camera, with starting times being and respectively, that is, the event images collected at the current moment and the previous moment.

[0069] Read (read from memory or storage) the center point coordinates of the moving obstacle in the two adjacent event images, the size information of the moving obstacle, and the distance (depth information) between the moving obstacle and the non-human body, and denote them as Then, within the time period, the moving distances X dis，n , Y dis，n of the moving obstacle in the horizontal and vertical directions, as well as the change in depth can be respectively expressed as:

[0070]

[0071]

[0072]

[0073] Then, the motion vector Expressed as:

[0074]

[0075] S32: Set the trigger condition for obstacle avoidance without human body according to the size information, motion vector of the target obstacle and the position of the target obstacle in the event image, and calculate the obstacle avoidance direction and movement distance.

[0076] As Figure 4 shown, set a warning area with two center point coordinates of in the event image with height n and width m. The height and width of warning area 1 are h warning1 and w warning1 pixels respectively, and the coordinates of the upper left vertex and the lower right vertex of the area are The height and width of warning area 2 are h warning2 and w warning2 pixels respectively, and the coordinates of the upper left vertex and the lower right vertex of the area are where h warning1 > h warning2 and w warning1 > w warning2 .

[0077] Then the trigger condition for obstacle avoidance is:

[0078] a. When the center point of the detected moving obstacle is in the event screen but outside warning area 1, and the depth Z < d 1 , trigger obstacle avoidance without human body;

[0079] b. When the center point of the detected moving obstacle is within warning area 1, that is and and when the depth change (indicating that the measured moving obstacle moves towards the non-human body and the depth gradually decreases) and the depth Z < d 2 , any of the following conditions is met to trigger obstacle avoidance without human body:

[0080] b1. If (indicating that the moving obstacle is on the left side of the non-human body) and (the first item of the motion vector of the moving obstacle) X dis，n > 0 (indicating that the moving obstacle moves from left to right);

[0081] b2. If (indicating that the moving obstacle is on the right side of the non-human body) and (the first item of the motion vector of the moving obstacle) X dis，n < 0 (indicating that the moving obstacle moves from right to left);

[0082] b3. If (Indicating that the moving obstacle is below the virtual human body) and (the second item of the motion vector of the moving obstacle) Y dis，n > 0 (indicating that the moving obstacle moves from bottom to top);

[0083] b4. If (Indicating that the moving obstacle is above the virtual human body) and (the first item of the motion vector of the moving obstacle) Y dis，n < 0 (indicating that the moving obstacle moves from top to bottom);

[0084] b5. When the center point of the detected moving obstacle is within the warning area 2, that is and at this time.

[0085] Wherein, d 1 and d 2 are respectively distance thresholds set for moving obstacles appearing outside and inside the warning area. Since the probability of collision between a moving obstacle and the virtual human body when the moving obstacle appears in a specific area increases in the above order, and the area with a greater collision probability requires an obstacle avoidance decision to be made at a farther distance, so d 1 < d 2 .

[0086] The present invention first determines whether the moving obstacle is outside the warning area. For the outside of the warning area, only a very small depth threshold is set, because the probability of collision between an object appearing at the edge of the screen and the virtual human body is very small, and obstacle avoidance measures only need to be taken when the object is very close to the virtual human body.

[0087] For a moving obstacle within the warning area, it is first determined whether the moving obstacle is getting closer to the virtual human body in depth. On the premise that the depth is gradually decreasing, it is further determined whether the moving obstacle is moving towards the virtual human body in the horizontal or vertical direction. When the above conditions are met, it indicates that the probability of collision between the object and the virtual human body is very high, and obstacle avoidance measures need to be taken at a relatively far distance from the virtual human body.

[0088] In addition to the above scenarios, if the moving obstacle moves directly towards the virtual human body, or although the moving obstacle does not move towards the virtual human body in the horizontal or vertical direction, but due to the physical size of the virtual human body itself, the moving obstacle moves towards the virtual human body from a small area directly in front of the virtual human body, in this case, obstacle avoidance measures also need to be taken at a relatively far distance.

[0089] If the obstacle avoidance is triggered at a certain moment, then the orthogonal vector of the motion vector of the moving obstacle is used as the obstacle avoidance direction of the virtual human body, that is For example, the obstacle avoidance direction vector can be expressed as:

[0090]

[0091] Alternatively, methods such as genetic algorithm, artificial potential field method, A* algorithm can also be used to set the obstacle avoidance direction.

[0092] After the human body obstacle avoidance is not triggered, the obstacle avoidance movement distance is the sum of the actual physical size of the non-human body and max(h, w), where max(h, w) is the larger value of the actual physical length and width of the moving obstacle.

[0093] The present invention combines a binocular event camera with the yolov5 object detection algorithm, can real-time complete the ranging and size measurement of moving target obstacle objects with unknown sizes, and deployed on non-human bodies such as unmanned aerial vehicles and unmanned vehicles can calculate the motion vector of the target obstacle object according to the above measurement results. Then, according to the further design of obstacle avoidance trigger conditions, obstacle avoidance direction and movement distance, the non-human body can complete the obstacle avoidance of fast-moving obstacles by virtue of the characteristics of high dynamic range, low latency and high dynamic performance of the event camera.

[0094] The present invention can effectively avoid unnecessary excessive obstacle avoidance maneuvering behaviors while ensuring reliable obstacle avoidance by calculating the motion vector of the moving obstacle object and setting two warning areas in the event image.

[0095] The present invention has a small calculation cost, can output an obstacle avoidance decision within a time of more than ten microseconds, is suitable for lightweight platforms, and can achieve autonomous obstacle avoidance by the on-board computing resources of the non-human body.

[0096] According to another embodiment of the present invention, an obstacle avoidance device for a non-human body is also disclosed, as Figure 5 shown, the device includes:

[0097] A binocular event camera 501 for real-time collecting event images of moving obstacles;

[0098] A first calculation unit 502 for calculating and saving the center point coordinates of the moving obstacle in the left event camera image and the right camera event image at each moment, as well as the size information and depth information of the moving obstacle at each moment;

[0099] A second calculation unit 503 for reading the center point coordinates of the moving obstacle in the continuous frame event images of any camera, as well as the size information and depth information of the moving obstacle, and calculating the motion vector of the target obstacle; and

[0100] An obstacle avoidance decision unit 504 for setting the non-human body obstacle avoidance trigger conditions according to the size information, motion vector of the target obstacle and the position of the target obstacle in the event image, and calculating the obstacle avoidance direction and movement distance.

[0101] According to still another embodiment of the present invention, an intelligent non-human body 600 is also disclosed, as Figure 6As shown, it includes a main body 601, a driving device 602 for driving the movement of the main body, and a control device 603. It also includes a binocular event camera 604 for collecting information on moving obstacles. The control device includes a processor 6031 and a memory 6032. A computer program is stored in the memory 6032, and the processor 6031 is used to execute the computer program to implement the obstacle avoidance method according to any one of the above solutions.

[0102] Figure 7 This is an embodiment of the present invention, taking a thrown ball as a moving obstacle. After the ball enters the warning area of the event screen and when the depth is less than the threshold, the algorithm gives an obstacle avoidance prompt, and at the same time outputs the movement vector of the moving obstacle and the obstacle avoidance direction vector of the absence of a human body (drone).

[0103] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. chips, or a combination of the above types of chips.

[0104] The memory can be a transient memory or a non-transient memory.

[0105] Although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments. Without departing from the concept of the present invention, any modification or equivalent replacement of the technical solutions of the embodiments of the present invention should not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An obstacle avoidance method for a humanoid robot without a human body, the humanoid robot without a human body including a binocular event camera, characterized in that, the method includes: Using the binocular event camera to collect event images of moving obstacles in real time; Calculating and saving the center point coordinates of the moving obstacles in the left event camera image and the right camera event image at each moment, as well as the size information and depth information of the moving obstacles at each moment; Reading the center point coordinates of the moving obstacles in the continuous frame event images of any camera, as well as the size information and depth information of the moving obstacles, and calculating the motion vector of the target obstacle; and Setting an obstacle avoidance trigger condition for the humanoid robot without a human body according to the size information, motion vector of the target obstacle, and the position of the target obstacle in the event image, and calculating the obstacle avoidance direction and motion distance; Wherein, two warning areas with different sizes centered on the center point of the event image are set, and the obstacle avoidance trigger condition is set according to the relative position between the center point of the moving obstacle and the two warning areas; When the center point of the moving obstacle is located in the event image, but outside the large warning area and the depth is less than the first threshold, the obstacle avoidance of the humanoid robot without a human body is triggered; When the center point of the moving obstacle is located within the large warning area, the depth change value is less than 0 and the depth is less than the second threshold, any of the following conditions is satisfied to trigger the obstacle avoidance of the humanoid robot without a human body: (i) the moving obstacle moves towards the humanoid robot without a human body in any dimension direction of the imaging plane; (ii) the center point of the moving obstacle is located within the small warning area.

2. The obstacle avoidance method according to claim 1, characterized in that, Using the binocular event camera to collect event images of moving obstacles in real time, and inputting the obtained left event image and right event image at each moment into a neural network respectively to obtain the target box vertex coordinates of the moving obstacle in its respective event image at the same moment.

3. The obstacle avoidance method according to claim 2, characterized in that, Using the target box vertex coordinates to calculate the center point coordinates of the moving obstacle in its respective event image, and calculating the length, width size and depth information of the moving obstacle based on the principle of similar triangles.

4. The obstacle avoidance method according to claim 1, characterized in that, After the obstacle avoidance is triggered, the orthogonal vector of the motion vector of the moving obstacle is used as the obstacle avoidance direction of the humanoid robot without a human body.

5. The obstacle avoidance method according to claim 4, characterized in that, After the obstacle avoidance is triggered, the obstacle avoidance motion distance is the sum of the actual physical size of the humanoid robot without a human body and the larger value of the length and width sizes of the moving obstacle.

6. An obstacle avoidance device for a humanoid robot without a human body, characterized in that, the device includes: A binocular event camera for collecting event images of moving obstacles in real time; A first calculation unit for calculating and saving the center point coordinates of the moving obstacles in the left event camera image and the right camera event image at each moment, as well as the size information and depth information of the moving obstacles at each moment; A second calculation unit for reading the center point coordinates of the moving obstacles in the continuous frame event images of any camera, as well as the size information and depth information of the moving obstacles, and calculating the motion vector of the target obstacle; and An obstacle avoidance decision-making unit, configured to set a no-human obstacle avoidance trigger condition according to the size information, motion vector of a target obstacle, and the position of the target obstacle in an event image, and calculate an obstacle avoidance direction and a motion distance; Among them, two warning areas with different sizes centered on the center point of the event image are set, and an obstacle avoidance trigger condition is set according to the relative position between the center point of the moving obstacle and the two warning areas; When the center point of the moving obstacle is located in the event image, but outside the large warning area and the depth is less than a first threshold, no-human obstacle avoidance is triggered; When the center point of the moving obstacle is located in the large warning area, the depth change value is less than 0 and the depth is less than a second threshold, no-human obstacle avoidance is triggered when any of the following conditions is met: (i) the moving obstacle moves in the direction of no human in any dimension direction of the imaging plane; (ii) the center point of the moving obstacle is located in the small warning area.

7. An intelligent no-human body, comprising a main body, a driving device for driving the main body to move, and a control device, characterized in that, it further comprises a binocular event camera for collecting information of moving obstacles, the control device comprises a processor and a memory, a computer program is stored in the memory, and the processor is configured to execute the computer program to implement the obstacle avoidance method according to any one of claims 1-5.

Images