WO2021012987A1

WO2021012987A1 - Object detection method, apparatus and system, and device

Info

Publication number: WO2021012987A1
Application number: PCT/CN2020/101827
Authority: WO
Inventors: 刘洪鑫; 苗振伟; 王兵
Original assignee: 阿里巴巴集团控股有限公司
Priority date: 2019-07-19
Filing date: 2020-07-14
Publication date: 2021-01-28
Also published as: CN112327308A; CN112327308B

Abstract

Disclosed are an object detection method, apparatus and system, and a device. The method comprises: determining at least one object at least according to three-dimensional coordinate data in road environment point cloud data (S101); acquiring a first reflection strength data set of the object (S103); and determining, according to the first reflection strength data set, whether the object is a real object (S105). By means of this processing method, reflection strength value information of a laser radar is effectively utilized, a falsely detected target is filtered out according to reflection strength value statistical information of a target object, and a laser radar point cloud is directly processed without relying on other sensor data, so as to realize the all-round removal of unreal objects by means of filtration, and thus, the accuracy of object detection can be effectively improved.

Description

Object detection method, device, system and equipment

This application claims the priority of the Chinese patent application with the application number 201910658506.4 and the invention title of "Object Detection Method, Device, System and Equipment" filed on July 19, 2019, the entire content of which is incorporated into this application by reference.

Technical field

This application relates to the field of unmanned driving technology, and in particular to an object detection method, device, system and equipment.

Background technique

Lidar is one of the most important sensors for unmanned driving systems, and has many advantages such as all-weather work, high precision, and three-dimensional measurement. However, its laser beam is more sensitive to targets such as rain, fog, dust, and automobile exhaust, and the measurement results may include these noise point clouds. Since conventional object detection methods may incorrectly identify these noise point clouds as obstacles such as vehicles and pedestrians, which may cause vehicles to perform additional avoidance actions and affect normal driving, how to filter out noise points caused by rain, fog, dust, exhaust gas, etc. The misdetection of lidar targets caused by clouds has become a research hotspot in this field.

At present, the processing process of a typical Lidar false detection target filtering method is as follows. First, collect the environmental point cloud data of the traffic road through the lidar installed on the vehicle, and take the environmental image of the traffic road through at least one camera installed on the vehicle; then, perform obstacle recognition based on the point cloud data and the environmental image respectively ; Finally, the lidar and the detection target of the camera are associated and fused to remove the false detection target.

However, in the process of implementing the present invention, the inventor found that the technical solution has at least the following problems: 1) The camera may not be able to provide 360-degree coverage of the detection results. If only the camera directly in front of the vehicle is working, the false detection targets in other areas are still Cannot be removed; 2) Because accurate internal and external parameter calibration is required to associate the lidar with the detection target of the camera, the fusion accuracy is low; 3) The image detection model involved in this solution needs to be trained from a large amount of annotation data , So the data cost is higher and the model training speed is lower. In summary, the prior art has the technical problem that it cannot effectively filter out the false detection target of lidar.

Summary of the invention

The present application provides an object detection method to solve the problem that the prior art cannot effectively identify the false detection target of the lidar. This application additionally provides object detection devices, systems and equipment, light source adjustment methods and equipment.

This application provides an object recognition method, including:

Determine at least one object at least according to the three-dimensional coordinate data in the road environment point cloud data;

Acquiring a first reflection intensity data set of the object;

According to the first reflection intensity data set, it is determined whether the object is a real object.

Optionally, the judging whether the object is a real object according to the first reflection intensity data set includes:

Determine, according to the first reflection intensity data set, a first point cloud quantity ratio corresponding to a plurality of preset reflection intensity values, to form a first point cloud quantity ratio vector of the object;

The object authenticity category prediction model determines the authenticity category of the object according to the first point cloud quantity ratio vector; the prediction model labels data from the first point cloud quantity ratio vector of the multiple objects and the object authenticity category Learned from the correspondence between.

Optional, also includes:

Based on the weighted cross-entropy loss function, the prediction model is learned from a plurality of the corresponding relationships; wherein, the first loss weight corresponding to the loss caused by misjudging a real object as an unreal object is greater than that of misjudging an unreal object It is judged as the second loss weight corresponding to the loss caused by the real object.

Optional, also includes:

Dividing the object into multiple sub-object blocks;

Acquiring a second reflection intensity data set of the sub-object block;

Determine, according to the second reflection intensity data set, a second point cloud quantity ratio corresponding to the plurality of preset reflection intensity values, to form a second point cloud quantity ratio vector of the sub-object block;

Using the prediction model to determine the authenticity category of the sub-object block according to the second point cloud quantity ratio vector;

According to the first point cloud data of the object and the authenticity category of each sub-object block, the second point cloud data of the object is determined.

Determine, according to the first reflection intensity data set, the number of first point clouds whose reflection intensity data of the object is greater than a threshold value of real object reflection intensity data;

According to the number of the first point cloud, it is determined whether the object is a real object.

Optionally, the judging whether the object is a real object according to the first point cloud quantity includes:

Acquiring a third point cloud quantity ratio between the first point cloud quantity and the second point cloud quantity of the object;

If the third point cloud quantity ratio is greater than the real object ratio threshold, the object is regarded as a real object.

Optional, also includes:

If the third point cloud quantity ratio is less than the ratio threshold, determining the number of reflection intensity values of the non-zero point cloud quantity of the object;

If the number of reflection intensity values is greater than the threshold value of the number of reflection intensity values, the object is regarded as a real object.

For each object, determine the number of reflection intensity values of the number of non-zero point clouds of the object;

Optionally, before the acquiring the first reflection intensity data set of the object, the method further includes:

Determining an object that affects the driving mode of the vehicle from the at least one object;

The judging whether the object is a real object according to the first reflection intensity data set includes:

For the object that affects the driving mode of the vehicle, it is determined whether the object is a real object according to the first reflection intensity data set.

Optionally, the objects that affect the driving mode of the vehicle include:

Objects located in front of the vehicle, objects located in the adjacent lane of the vehicle.

Optional, also includes:

If the number of unreal objects determined according to the first reflection intensity data set satisfies the alarm condition, the alarm information is displayed, so that the user controls the vehicle to enter the non-unmanned driving mode according to the alarm information.

Optional, also includes:

Regarding the first real object determined according to the first reflection intensity data set, it is determined whether the first real object is a second real object according to historical object recognition data and/or non-laser reflection intensity data.

Optional, also includes:

For the first non-real object determined according to the first reflection intensity data set, it is determined whether the first non-real object is a second non-real object according to historical object recognition data and/or non-laser reflection intensity data.

This application also provides an object detection method, including:

Determine at least one object according to the road environment point cloud data;

Dividing the object into multiple sub-object blocks;

Judging whether the sub-object block is a real object block according to the reflection intensity data set of the sub-object block;

According to the first point cloud data of the object and the above judgment result, the second point cloud data of the object is determined.

Optionally, the judging whether the sub-object block is a real object block according to the reflection intensity data set includes:

According to the reflection intensity data set, determine the ratio of the number of point clouds respectively corresponding to a plurality of preset reflection intensity values, and form the ratio vector of the number of point clouds of the sub-object block;

Through the object authenticity category prediction model, the authenticity category of the sub-object block is determined according to the point cloud quantity ratio vector.

This application also provides an object detection device, including:

The object prediction unit is configured to determine at least one object at least according to the three-dimensional coordinate data in the road environment point cloud data;

A data acquisition unit for acquiring a first reflection intensity data set of the object;

The real object determining unit is configured to determine whether the object is a real object according to the first reflection intensity data set.

This application also provides an object detection device, including:

Processor; and

The memory is used to store the program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: at least according to the three-dimensional coordinate data in the road environment point cloud data, determine at least An object; obtain a first reflection intensity data set of the object; determine whether the object is a real object according to the first reflection intensity data set.

Optionally, the device includes: a vehicle or a road test sensing device.

This application also provides an object detection device, including:

The object prediction unit is used to determine at least one object according to the road environment point cloud data;

An object segmentation unit for segmenting the object into multiple sub-object blocks;

The real object block determining unit is configured to determine whether the sub object block is a real object block according to the reflection intensity data set of the sub object block;

The object determining unit is used for the object to determine the second point cloud data of the object according to the first point cloud data of the object and the above judgment result.

This application also provides an object detection device, including:

Processor; and

The memory is used to store a program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: determine at least one object according to the road environment point cloud data; The object is divided into multiple sub-object blocks; according to the reflection intensity data set of the sub-object blocks, it is judged whether the sub-object blocks are real object blocks; according to the first point cloud data of the object and the above judgment result, the The second point cloud data of the object.

This application also provides an object detection system, including:

A terminal device for collecting road environment point cloud data of the terminal device, sending an object detection request for the road environment point cloud data to the server; and receiving real object point cloud data information returned by the server;

The server is configured to receive the request, determine at least one object based on at least the three-dimensional coordinate data in the road environment point cloud data; obtain the reflection intensity data set of the object; determine the reflection intensity data set according to the reflection intensity data set. Whether the object is a real object; sending back the point cloud data information of the real object to the terminal device.

This application also provides an object detection system, including:

The terminal device is configured to collect the road environment point cloud data of the terminal device, send an object detection request for the road environment point cloud data to the server; and receive the object point cloud data returned by the server;

The server is configured to receive the request, determine at least one object based on the road environment point cloud data; divide the object into multiple sub-object blocks; determine the sub-object based on the reflection intensity data set of the sub-object blocks Whether the block is a real object block; determine the second point cloud data of the object according to the first point cloud data of the object and the above judgment result; send the second point cloud data back to the terminal device.

This application also provides an object detection method, including:

Collect road environment point cloud data;

Sending an object detection request for the road environment point cloud data to the server;

Receiving real object point cloud data information returned by the server.

This application also provides an object detection method, including:

Receive object detection requests for road environment point cloud data;

At least determining at least one object according to the three-dimensional coordinate data in the road environment point cloud data;

Acquiring a reflection intensity data set of the object;

Judging whether the object is a real object according to the reflection intensity data set;

Send back the point cloud data information of the real object to the requesting party.

This application also provides an object detection device, including:

Data collection unit, used to collect road environment point cloud data;

A request sending unit, configured to send an object detection request for the road environment point cloud data to the server;

The data receiving unit is configured to receive real object point cloud data information returned by the server.

This application also provides an object detection device, including:

Processor; and

The memory is used to store the program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: collect road environment point cloud data; An object detection request for environmental point cloud data; receiving real object point cloud data information returned by the server.

This application also provides an object detection device, and this application also provides:

The request receiving unit is configured to receive an object detection request for road environment point cloud data;

An object prediction unit, configured to determine at least one object at least according to the three-dimensional coordinate data in the road environment point cloud data;

A data acquisition unit for acquiring the reflection intensity data set of the object;

A real object determining unit, configured to determine whether the object is a real object according to the reflection intensity data set;

The data return unit is used to return real object point cloud data information to the requesting party.

This application also provides an object detection device, including:

Processor; and

The memory is used to store the program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: receiving an object detection request for road environment point cloud data; According to the three-dimensional coordinate data in the road environment point cloud data, at least one object is determined; the reflection intensity data set of the object is acquired; the reflection intensity data set is used to determine whether the object is a real object; the real object is sent back to the requesting party Point cloud data information.

This application also provides an object detection method, including:

Collect road environment point cloud data;

Receiving the object point cloud data information returned by the server.

This application also provides an object detection method, including:

Receive object detection requests for road environment point cloud data;

Dividing the object into multiple sub-object blocks;

Determine the second point cloud data of the object according to the first point cloud data of the object and the foregoing judgment result;

Send the second point cloud data back to the requesting party.

This application also provides an object detection device, including:

Data collection unit, used to collect road environment point cloud data;

The data receiving unit is used to receive the object point cloud data information returned by the server.

This application also provides an object detection device, including:

Processor; and

The memory is used to store the program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: collect road environment point cloud data; Object detection request for environmental point cloud data; receiving object point cloud data information returned by the server.

This application also provides an object detection device, including:

An object determining unit, configured to determine the second point cloud data of the object according to the first point cloud data of the object and the above judgment result;

The data return unit is configured to return the second point cloud data to the requesting party.

This application also provides an object detection device, including:

Processor; and

The memory is used to store the program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: receiving an object detection request for the road environment point cloud data; according to the road environment Point cloud data, determine at least one object; divide the object into multiple sub-object blocks; determine whether the sub-object block is a real object block according to the reflection intensity data set of the sub-object block; The point cloud data and the above judgment result determine the second point cloud data of the object; the second point cloud data is sent back to the requesting party.

This application also provides a light source adjustment method, including:

Determine at least one object based on road environment data;

Determining the number of non-real objects in the at least one object;

According to the number of the unreal objects, the working mode of the incident light source of the object is adjusted.

Optionally, the adjusting the working mode of the incident light source of the object according to the number of non-real objects includes:

If the number of unreal objects meets the light source adjustment condition, the working mode of the incident light source of the object is adjusted.

Optionally, the light source adjustment conditions include:

The number of non-real objects is greater than the number threshold, and/or the ratio between the number of non-real objects and the number of the at least one object is greater than the ratio threshold.

Optionally, the adjusting the working mode of the incident light source of the object adopts at least one of the following modes:

Turning on or off the lights of a first vehicle, the first vehicle including a vehicle that executes the method;

Adjust the brightness of the lights of the first vehicle.

Optionally, the adjusting the working mode of the incident light source of the object includes:

Determining a target second vehicle, where the target second vehicle includes a vehicle adjacent to the first vehicle;

The first instruction information for turning on or off the lights of the target second vehicle is sent to the target second vehicle through the signal sending device, so that the target second vehicle receives the first instruction information through the signal receiving device, and according to the first instruction information Turn on or off the lights of the target second vehicle.

The signal sending device sends the second instruction information for adjusting the brightness of the lamp of the target second vehicle to the target second vehicle, so that the target second vehicle receives the second instruction information through the signal receiving device, and adjusts according to the second instruction information Target the brightness of the lights of the second vehicle.

Determining a target street light, where the target street light includes a vehicle adjacent to the first vehicle;

The signal sending device sends third instruction information for turning on or off the target street light to the target street light, so that the target street light receives the third instruction information through the signal receiving device, and turns on or off the target street light according to the third instruction information.

The signal sending device sends the fourth instruction information for adjusting the brightness of the street light to the target street light, so that the target street light receives the fourth instruction information through the signal receiving device, and adjusts the brightness of the street light according to the fourth instruction information.

The application also provides a light source adjustment device, including:

The object prediction unit is used to determine at least one object according to road environment data;

A unit for determining the number of non-real objects, configured to determine the number of non-real objects in the at least one object;

The light source adjusting unit is used to adjust the working mode of the incident light source of the object according to the number of the unreal object.

This application also provides a vehicle, including:

Processor; and

The memory is used to store a program for realizing the light source adjustment method. After the device is powered on and runs the light source adjustment method program through the processor, the following steps are executed: determine at least one object according to road environment data; determine the at least one The number of non-real objects in the object; according to the number of non-real objects, the working mode of the incident light source of the object is adjusted.

This application also provides a vehicle, including:

Signal receiving device;

Processor; and

The memory is used to store a program for realizing the light source adjustment method. After the device is powered on and runs the light source adjustment method program through the processor, the following steps are executed: receiving light source adjustment instruction information; adjusting the light source adjustment method according to the instruction information How the light source attached to the vehicle works.

This application also provides a light source adjustment method, including:

Receive light source adjustment instruction information;

According to the instruction information, the working mode of the light source attached to the vehicle is adjusted.

This application also provides a street lamp, including:

Signal receiving device;

Processor; and

The memory is used to store a program for realizing the light source adjustment method. After the device is powered on and runs the light source adjustment method program through the processor, the following steps are executed: receiving light source adjustment instruction information; adjusting the light source adjustment method according to the instruction information How street lights work.

This application also provides a light source adjustment method, including:

Receive light source adjustment instruction information;

According to the instruction information, adjust the working mode of the street lamp.

This application also provides an object detection method, including:

Determine at least one object based on road environment data;

Acquiring sound wave reflection intensity data of the object;

According to the sound wave reflection intensity data, it is determined whether the object is a real object.

Optionally, the acquiring sound wave reflection intensity data of the object includes:

Collect sound wave reflection intensity data of the road environment through ultrasonic sensors;

Determine the sound source location information according to the sound wave reflection intensity data of the road environment;

According to the sound source position information and the three-dimensional coordinate data of the object, the sound wave reflection intensity data of the object is determined.

This application also provides an object detection device, including:

A data acquisition unit for acquiring sound wave reflection intensity data of the object;

The real object determining unit is configured to determine whether the object is a real object according to the sound wave reflection intensity data.

This application also provides an object detection device, including:

Processor; and

The present application also provides a computer-readable storage medium having instructions stored in the computer-readable storage medium, which when run on a computer, cause the computer to execute the above-mentioned various methods.

The present application also provides a computer program product including instructions, which when run on a computer, causes the computer to execute the above-mentioned various methods.

Compared with the prior art, this application has the following advantages:

According to the object detection method provided by the embodiment of the present application, at least one object is determined according to at least three-dimensional coordinate data in the road environment point cloud data; a first reflection intensity data set of the object is acquired; according to the first reflection intensity data set , To determine whether the object is a real object; this processing method makes effective use of the reflection intensity value information of the lidar, and performs false detection target filtering based on the reflection intensity value statistical information of the target object, avoiding relying on other sensor data , But directly process the lidar point cloud to achieve a full range of unreal object filtering; therefore, it can effectively improve the accuracy of object detection.

The object detection method provided by the embodiments of the present application determines at least one object according to the road environment point cloud data; divides the object into a plurality of sub-object blocks; determines the sub-object block according to the reflection intensity data set of the sub-object blocks Whether the object block is a real object block; determine the second point cloud data of the object according to the first point cloud data of the object and the above judgment result; this processing method makes it possible to determine a more accurate object bounding box; Therefore, the accuracy of object detection can be effectively improved.

The light source adjustment method provided by the embodiment of the present application determines at least one object according to road environment data; determines the number of non-real objects in the at least one object; and adjusts the incident light source of the object according to the number of non-real objects Working method; this processing method makes it possible to improve the environment of the vehicle driving road in real time, so as to collect higher-quality road environment perception data; therefore, it can effectively improve the safety of unmanned driving.

The object detection method provided by the embodiments of the present application determines at least one object according to road environment data; obtains the sound wave reflection intensity data of the object; and determines whether the object is a real object according to the sound wave reflection intensity data; The processing method makes effective use of the sound wave reflection intensity value information, and filters out false targets based on the sound wave reflection intensity data of the target object to achieve a full range of unreal object filtering; therefore, it can effectively improve the accuracy of object detection .

Description of the drawings

Fig. 1 is a flowchart of an embodiment of an object detection method provided by the present application;

FIG. 2 is an image of a road environment in a water fog situation in an embodiment of the object detection method provided by the present application;

FIG. 3 is a schematic diagram of point cloud data of an embodiment of the object detection method provided by the present application;

FIG. 4 is a specific flowchart of an embodiment of the object detection method provided by the present application;

5 is a schematic diagram of the prediction model network structure of an embodiment of the object detection method provided by the present application;

6 is a schematic diagram of a bounding box of an embodiment of the object detection method provided by the present application;

FIG. 7 is a histogram of statistical information distribution of an embodiment of the object detection method provided by the present application;

FIG. 8 is an image of a road environment in the case of dust in the embodiment of the object detection method provided by the present application;

FIG. 9 is another schematic diagram of cloud data of an embodiment of the object detection method provided by the present application;

10 is another statistical information distribution histogram of the embodiment of the object detection method provided by this application;

FIG. 11 is another statistical information distribution histogram of the embodiment of the object detection method provided by the present application;

FIG. 12 is a flowchart of an embodiment of the object detection method provided by the present application;

FIG. 13 is a schematic diagram of an embodiment of the object detection system provided by the present application;

FIG. 14 is a schematic diagram of an embodiment of the object detection system provided by the present application;

15 is a flowchart of an embodiment of a light source adjustment method provided by the present application;

FIG. 16 is a flowchart of an embodiment of the object detection method provided by the present application.

Detailed ways

In the following description, many specific details are explained in order to fully understand this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar promotion without violating the connotation of this application. Therefore, this application is not limited by the specific implementation disclosed below.

In this application, object detection methods, devices, systems and equipment, and light source adjustment methods and equipment are provided. In the following embodiments, each solution will be described in detail.

First embodiment

Please refer to FIG. 1, which is a flowchart of an embodiment of an object detection method provided by this application. The execution subject of the method includes but is not limited to an unmanned vehicle, and may also be a road test sensing device and other devices. An object detection method provided by this application includes:

Step S101: Determine at least one object at least according to the three-dimensional coordinate data in the road environment point cloud data.

In the method provided by the embodiments of the present application, during the driving process of the vehicle, the three-dimensional space scanning device installed on the vehicle can obtain the spatial coordinates of each sampling point on the surface of the environmental space object of the vehicle driving road to obtain a collection of points Point data is called point cloud data of road environment. Through the road environment point cloud data, the scanned object surface is recorded in the form of points. Each point contains three-dimensional coordinate data and reflection intensity data (Intensity), and some may contain color data (RGB). With point cloud data, the target space can be expressed in the same spatial reference system.

The three-dimensional space scanning device may be a Lidar (Light Detection And Ranging, Lidar), which performs laser detection and measurement through a laser scanning method to obtain information about obstacles in the surrounding environment, such as buildings, trees, people, vehicles, etc. The measured data is the discrete point representation of the Digital Surface Model (DSM). In specific implementation, 16-line, 32-line, 64-line and other multi-line lidars can be used. Radars with different laser beam numbers have different frame rates for collecting point cloud data. For example, 16 and 32 lines generally collect 10 frames per second. Cloud data. The three-dimensional space scanning device may also be equipment such as a three-dimensional laser scanner or a photographic scanner.

After the vehicle in this embodiment collects road environment point cloud data through the three-dimensional space scanning device, at least one object can be determined according to the three-dimensional coordinate data in the road environment point cloud data of the current frame.

The road environment point cloud data may include point cloud data of various real objects in the road environment space, and these objects may be trees, buildings, pedestrians and vehicles on the road, and so on. The road environment point cloud data may also include point cloud data of substances such as water mist, dust, exhaust gas, etc. in the road environment space. Such point cloud data makes the object determined according to the road environment point cloud data may also be non-real objects, such as Non-real objects represented by water mist point cloud data clusters, non-real objects represented by exhaust point cloud data clusters, or non-real objects represented by dust point cloud data clusters, etc. Therefore, the object determined in step S101 may be a real object or an unreal object.

In an example, at least one object is identified from the road environment point cloud data through an object detection model. In specific implementation, after a frame of environmental point cloud data is obtained by scanning the lidar on the vehicle, the environmental point cloud data can be transmitted to the object detection model, through which the categories of various objects (such as vehicles, pedestrians, trees, etc.) can be detected Etc.) and object point cloud data information. The object point cloud data information may be three-dimensional position data, such as vertex coordinate data of a rectangular cube bounding box of the object, and so on.

In specific implementation, the object detection model can use the RefineDet method based on deep learning. This method uses the fast running speed of single-stage methods such as SSD for reference, and combines two-stage methods such as Faster R-CNN, so it has The advantage of high object detection accuracy. When this method detects the object point cloud data in the environmental point cloud data, it obtains the bounding box coordinates of the object.

In another example, the point cloud data clustering method can also be used to cluster the point cloud data that are closer together to form objects based on the three-dimensional coordinate data in the road environment point cloud data. This type of object is a point cloud clustering the result of.

Figure 2 shows the image collected by the camera when driving on a rainy day. Correspondingly, Figure 3 shows the three-dimensional point cloud data collected by the lidar at this moment. In the laser point cloud image, the points with different gray levels are the laser point cloud. The frame formed by the line is the object bounding frame. It can be seen from FIG. 3 that in this embodiment, 5 obstacle clusters are generated through step S101, which are numbered 1, 2, 3, 4, and 5 respectively. The clusters numbered 1, 3, and 5 are real objects (vehicles), and the clusters numbered 2 and 4 are false detections caused by water mist splashed by vehicles, that is, non-real objects.

After determining the object, you can proceed to the next step to obtain the first reflection intensity data set of the object.

Step S103: Obtain a first reflection intensity data set of the object.

The road environment point cloud data includes not only the three-dimensional coordinate data of each object point, but also the reflection intensity data of each object point. The first reflection intensity data set of the object includes reflection intensity data of various points of the object.

In an example, the object is an object recognized by an object prediction model, and the output data of the object prediction model may include vertex coordinate data of a bounding box of the object. Step S103 may include the following sub-steps: 1) Obtain the point cloud data of the object from the road environment point cloud data according to the vertex coordinate data of the bounding box of the object; 2) Obtain the reflection intensity data of the object from the point cloud data of the object.

In another example, the object is an object determined by a point cloud data clustering algorithm, and the input data of the algorithm may include the point cloud data of each object, so the object can be obtained directly from the object point cloud data obtained in step S101 Reflected intensity data.

After acquiring the first reflection intensity data set of the object, it is possible to proceed to the next step to filter out the real objects in the at least one object according to the reflection intensity data in the object point cloud data.

Step S105: Determine whether the object is a real object according to the first reflection intensity data set.

The at least one object determined in step S101 may include an unreal object (misdetection target). In this step, the laser reflection intensity data of the object is used, and the misdetection target is identified according to the statistical information of the reflection intensity data, so as to determine the The real object in the at least one object is used as the final object detection result.

Three implementation manners that can be adopted in step S105 are given below and described.

Manner 1. As shown in FIG. 4, step S105 in this embodiment may include the following sub-steps:

Step S1051: According to the first reflection intensity data set, determine a first point cloud quantity ratio corresponding to a plurality of preset reflection intensity values, and form a first point cloud quantity ratio vector of the object.

The multiple preset reflection intensity values may include 256 reflection intensity values ranging from 0 to 255, or may only include partial reflection intensity values, such as 200. Correspondingly, the dimension of the first point cloud quantity ratio vector is the same as the numerical quantity of the plurality of preset reflection intensity values.

The first point cloud number ratio is the ratio between the number of object points whose reflection intensity data included in the object is a certain preset reflection intensity value and the number of all object points included in the object. For example, if the number of points included in an object is 100 and the number of points corresponding to the reflection intensity value of 15 is 20, the ratio of the number of first point clouds corresponding to the reflection intensity value of 15 is 0.2; the number of points corresponding to the reflection intensity value of 20 is 10, the ratio of the number of first point clouds corresponding to the reflection intensity value of 20 is 0.1, and so on. The dimension of the first point cloud quantity ratio vector in this embodiment is 256 dimensions.

Step S1053: Determine the authenticity class of the object according to the first point cloud quantity ratio vector through the object authenticity class prediction model.

As shown in FIG. 5, the network structure of the prediction model may be a two-classifier neural network, including an input layer (for example, a 256-dimensional input layer), a hidden layer (for example, a 512-dimensional hidden layer), and an output layer (for example, 2D output layer). The input data of the model is the first point cloud quantity ratio vector, and the output data is the authenticity category of the object, which can be a real object or an unreal object.

The prediction model can be learned from the correspondence relationship between the first point cloud quantity ratio vector of multiple objects and the object authenticity category annotation data. Table 1 shows the training data of the prediction model.

训练样本标识Training sample identification	第一点云数量比值向量(256维)The first point cloud number ratio vector (256 dimensions)	物体真实性类别Object authenticity category
11	(0.01,0.05,…,0)(0.01,0.05,…,0)	真实物体(正样本)Real object (positive sample)
22	(0.9,0.09,…,0)(0.9,0.09,…,0)	非真实物体(负样本)Non-real objects (negative samples)
33	(0.01,0.008,…,0)(0.01,0.008,…,0)	真实物体(正样本)Real object (positive sample)
……	……	……
106,000106,000	(0.87,0.1,…,0)(0.87,0.1,…,0)	非真实物体(负样本)Non-real objects (negative samples)

Table 1. The training data set of the prediction model

In this embodiment, 106,000 training samples are included, where the number of positive samples is much larger than the number of negative samples, the number of positive samples is about 100,000, and the number of negative samples is about 6,000. In specific implementation, a model training method of multiple iterations can be used to obtain model parameters with higher accuracy. For example, the sample size of each iteration training is 10,000, and the samples of different times of training can partially overlap. Due to the large number of positive samples, the positive sample for each iteration training can be determined by positive sample downsampling, so that the number of positive/negative samples is more balanced, avoiding the model from over-fitting the real object, and directly adding the object under test Divided into real objects.

It should be noted that the prediction model is a model used to identify the authenticity category of the measured object. When the model misjudges an unreal object as a real object, it only causes the vehicle to perform additional avoidance actions, affecting normal driving. But it will not cause driving danger. However, if the model misjudges real objects as non-real objects, it will lead to traffic accidents. Since the misjudgment of a real object as an unreal object will lead to catastrophic results, it is impossible to accurately evaluate the performance of the model through the conventional model global accuracy index, which places higher requirements on the accuracy of the model.

In order to evaluate the performance of the prediction model more effectively, this embodiment trains the prediction model in the following manner: based on a weighted cross entropy loss function (weighted cross entropy), the prediction model is obtained by learning from multiple correspondences. Among them, the first loss weight corresponding to the loss caused by the misjudgment of the real object as an unreal object is greater than the second loss weight corresponding to the loss caused by the misjudgment of the unreal object as the real object. For example, the first loss weight is the first loss weight. 2. Loss weight 5 times and so on. The loss function of this embodiment is as follows:

In the above formula, x is the neuron vector (two-dimensional) output by the model, class is the true category (0 or 1) of the training sample; the weight vector is the weight value for different categories, for example, set to [1,5] to indicate "The loss of misclassification of positive samples is 5 times that of negative samples; the value of j is 0 or 1.

In the method provided by the embodiment of the present application, the prediction model is learned from multiple correspondences based on the weighted cross-entropy loss function, so that the loss value of the positive sample is misjudged as the negative sample is amplified, and the loss value is punished. Therefore, not only can the model's recall of real objects be effectively improved, but also the model's misjudgment rate of real objects can be effectively reduced. Experiments have proved that with the method provided in this embodiment, the recall rate can reach 99.9%, and the accuracy rate can reach 93.2%.

In this embodiment, the method may further include the following steps: 1) dividing the object into a plurality of sub-object blocks; 2) acquiring a second reflection intensity data set of the sub-object blocks; 3) according to the first A second reflection intensity data set, determining the ratio of the second point cloud quantity corresponding to the plurality of preset reflection intensity values, and forming the second point cloud quantity ratio vector of the sub-object block; 4) through the prediction model, Determine the authenticity category of the sub-object block according to the second point cloud quantity ratio vector; 5) determine the second point cloud data of the object and the authenticity category of each sub-object block Point cloud data, for example, if the sub-object block is an unreal object, the point cloud data of the sub-object block is cleared from the first point cloud data of the object, so that the second point cloud data of the object does not include Point cloud data of non-real sub-object blocks.

As shown in Fig. 6, a) in Fig. 6 shows the actual bounding box of the object, and b) in Fig. 6 shows the enlargement of the tail part including dust and rain caused by the object detection model. The bounding box (the bounding box determined in step S101), Figure 6 c) shows a number of small grids (sub-object blocks) included in the enlarged bounding box, that is, grid-level classification, d in Figure 6 ) Shows the bounding box determined through the above steps 1-5, which is closer to the actual bounding box, that is, the accurate bounding box.

The foregoing step 2 corresponds to step S103, step 3 corresponds to step S1051, and step 4 corresponds to step S1053. Since the principles of the two are the same, they will not be repeated here. When obtaining a more accurate object bounding box based on the object bounding box determined in step S101, the prediction model described in step S1053 can be directly used.

In the method provided by the embodiment of the present application, through the above steps 1-5, a more accurate object bounding box is obtained according to the object bounding box determined in step S101, and therefore, the accuracy of object detection can be effectively improved.

The first implementation of step S105 has been described above.

Manner 2: Step S105 may include the following sub-steps:

Step S1051': According to the first reflection intensity data set, determine the first point cloud number of which the reflection intensity data of the object is greater than the threshold value of the reflection intensity data of the real object.

The gray value of the laser point cloud in Figure 3 reflects its reflection intensity value. The reflection intensity value of the point with the lowest gray value is 0, the reflection intensity value of the point with the middle gray value is 1, and the point with the highest gray value is The reflection intensity value is greater than or equal to 2. Fig. 7 shows a histogram of the distribution of laser point cloud reflection intensity values belonging to the 5 clusters in Fig. 3, from top to bottom are the distribution of point cloud reflection intensity values of clusters numbered 1, 2, 3, 4, and 5. It can be seen from Figure 7 that the reflection intensity values of water mist objects are mostly concentrated in 0 and 1, while real objects (such as vehicles) have a relatively wide distribution of intensity values, and there is a larger proportion of point clouds in the range of reflection intensity greater than or equal to 2. . Similarly, in the case of dust and exhaust, the distribution of laser point clouds has similar properties. Figure 8 is a night scene with dust, and Figure 9 is the corresponding laser point cloud. Figure 9 has 8 clusters numbered 1, 2, 3, 4, 5, 6, 7, and 8. Figure 10 shows the intensity distribution of the point cloud corresponding to the 8 clusters. The point cloud intensity of dust and exhaust gas is also concentrated in two reflection intensity values of 0 and 1.

In order to further analyze the intensity distribution of the laser point cloud under vehicles, rain, fog, and dust, the inventors calculated the intensity distribution of clusters of rain, fog, and dust in 1000 groups of vehicles randomly sampled. The point cloud intensity distribution is shown in Figure 11. This distribution diagram shows that the intensity distribution of point cloud under dust, rain and fog is very different from the intensity distribution of vehicles. To this end, the inventor proposes a rule-based method to filter out the effects of dust, rain, fog, and exhaust gas, which includes step S1051' and step S1053'.

The reflection intensity data threshold value of the real object includes the lower limit value of the reflection intensity of the real object. If the threshold value is set to 2, the number of points with the reflection intensity value of the object greater than 2 can be obtained through this step, that is, the first The number of point clouds.

Step S1053': Determine whether the object is a real object according to the number of the first point cloud.

In an example, the authenticity category of the object can be determined directly according to the number of the first point cloud. For example, the object with the first point cloud number greater than 50 is regarded as the real object, and the object with the first point cloud number less than or equal to 50 As an unreal object.

In another example, step S1053' may include the following sub-steps: 1) Obtain the third point cloud number ratio between the first point cloud number and the second point cloud number of the object; 2) If the The third point cloud quantity ratio is greater than the real object ratio threshold, and the object is regarded as a real object.

Among them, the second point cloud quantity is the quantity of all points of the object. The real object ratio threshold includes the lower limit value of the third point cloud quantity ratio of the real object. If the threshold is set to 0.8 and the third point cloud quantity ratio of the measured object is 0.85, the measured object is regarded as the real Object: The third point cloud quantity ratio of the measured object is 0.79, and the measured object is regarded as an unreal object.

With this processing method, the number of first point clouds is normalized to avoid misjudgement of larger-size objects as real objects and smaller-size objects as unreal objects; therefore, the accuracy of object detection can be effectively improved rate.

In specific implementation, the following steps may be included after step S1053':

Step S1055': If the third point cloud quantity ratio is less than the ratio threshold, determine the reflection intensity value quantity of the non-zero point cloud quantity of the object.

In the case that the third point cloud number ratio is less than the ratio threshold, the number of reflection intensity values of the non-zero point cloud number of the object can be determined. For example, the reflection intensity of a water mist object is concentrated on two values of 0 and 1. Among the 256 reflection intensity values from 0 to 255, only the number of points with the two reflection intensity values of 0 and 1 is greater than 0, and the reflection intensity The number of points with values from 2 to 255 is equal to 0, then the number of reflection intensity values of the non-zero point cloud number of the water mist object is 2, including the two reflection intensity values of 0 and 1.

Step S1057': If the number of reflection intensity values is greater than the threshold value of the number of reflection intensity values, the object is regarded as a real object.

The threshold for the number of reflection intensity values includes the lower limit of the number of reflection intensity values for real objects. If the threshold is set to 3 and the number of reflection intensity values for the measured object is 10, the measured object is regarded as a real object; The number of reflected intensity values of the measured object is 2, and the measured object is regarded as an unreal object.

The second implementation of step S105 has been described above.

Manner 3: Step S105 may include the following sub-steps: 1) For each object, determine the number of reflection intensity values for the number of non-zero point clouds of the object; 2) If the number of reflection intensity values is greater than the threshold value of the reflection intensity value, then The object is regarded as a real object.

The third method is similar to the above step S1055' and step S1057', and will not be repeated here.

It should be noted that when the implementation of the second or third method is adopted, since the authenticity of the object is determined directly based on the statistical information of the reflection intensity value, there is no need to learn samples, which can effectively reduce the data cost and save the computing resources and storage resources. , And improve the recognition speed.

In an example, before step S103, the method provided in the embodiment of the present application may further include the following step: determining an object that affects the driving mode of the vehicle from the at least one object.

The object that affects the driving mode of the vehicle refers to the object that affects the driving mode of the vehicle. For example, an object located in front of the vehicle may cause the vehicle to decelerate or accelerate, and an object located in the adjacent lane of the vehicle may affect the timing of the vehicle changing lane, etc. , These objects are the objects that affect the way the vehicle travels. Objects located behind the vehicle usually do not affect the driving mode of the vehicle, and these objects are the objects that do not affect the driving mode of the vehicle.

The method provided by the embodiment of the present application determines the object that affects the driving mode of the vehicle from the at least one object before step S103, and executes step S105 only for the object that affects the driving mode of the vehicle, so that the number of discriminated objects is reduced. On the premise of ensuring safe driving, calculation costs can be effectively saved.

In another example, before step S103, the method provided in this embodiment of the present application may further include the following step: if the number of unreal objects determined according to the first reflection intensity data set meets the alarm condition, display alarm information to Allows the user to control the vehicle to enter the non-unmanned driving mode according to the alarm information.

The alarm condition may be that the number of non-real objects is greater than the number threshold. For example, the number threshold is set to 3. If the number of non-real objects exceeds the threshold, it means that there are too many non-real objects and the current driving environment is poor (such as large water fog, Severe dust, etc.), it is necessary to manually take over the driving of the vehicle, so an alarm message is displayed to remind the user (driver) to control the driving of the vehicle through manual driving mode. The alarm condition may also be that the ratio of the number of non-real objects to the number of the at least one object is greater than a ratio threshold (such as 20%), or the number of non-real objects is greater than the number threshold, and the number of non-real objects is greater than the at least one object. The ratio of the number of objects is greater than the ratio threshold.

The display mode of the alarm information may be a voice mode, or may be directly displayed on a display screen in the vehicle, and so on.

The method provided by the embodiment of this application determines whether the number of non-real objects meets the alarm condition after step S105, and if the result of the judgment is yes, the alarm information is displayed, so that the user controls the vehicle to enter the non-unmanned driving mode according to the alarm information ; This approach can effectively improve the driving safety of the vehicle.

In another example, the method provided by the embodiment of the present application may further include the following steps: for the first real object determined according to the first reflection intensity data set, according to historical object identification data and/or non-laser reflection intensity data, Determine whether the first real object is a second real object.

The historical object recognition data may include object information recognized according to point cloud data of several frames before the current frame. For example, if it is determined that there is no car in front of the vehicle for a long time according to the object recognition data and object position data of the first 10 frames, the first real object determined in step S105 is likely to be an unreal object such as a mist object or a dust object.

The non-laser reflection intensity data may be sound wave reflection intensity data, or a road environment image or the like.

To determine whether the first real object is the second real object based on the sound wave reflection intensity data, the following processing steps may be included: 1) Acquire sound wave reflection intensity data of the road environment through an ultrasonic sensor; 2) According to the sound wave reflection intensity data , Determine the sound source position information; 3) determine whether the first real object is a second real object according to the sound source position information and the three-dimensional coordinate data of the first real object.

For example, if the distance between the sound source location information and the three-dimensional coordinate data of the first real object is less than the distance threshold, it means that the reflected sound wave originates from the first real object. Therefore, the object is a real object and is taken as the first real object. 2. Real object; if the distance is greater than or equal to the distance threshold, it means that the sound wave does not originate from the first real object, and the ultrasonic wave emitted by the ultrasonic sensor does not reflect back, but penetrates the first real object. Therefore, the object Not a real object.

To judge whether the first real object is the second real object according to the road environment image, the following processing process can be adopted: collecting the environmental point cloud data of the traffic road through the laser radar installed on the vehicle, and at least A camera captures the environmental image of the traffic road; then, obstacle recognition is performed based on the point cloud data and the environmental image respectively; finally, the lidar and the detection target of the camera are associated and fused to remove the false target.

The method provided by the embodiment of the present application can effectively improve the accuracy of object recognition through the step of determining whether the first real object is the second real object. If it is determined that the first real object determined in step S105 is not a real object, This can effectively reduce the vehicle's additional avoidance actions and avoid affecting normal driving.

In specific implementation, if the number of unreal objects determined according to the first reflection intensity data set meets and re-identify conditions, then for the first real object determined according to the first reflection intensity data set, the historical object identification Data and/or non-laser reflection intensity data to determine whether the first real object is a second real object.

The re-identification condition may be that the number of non-real objects is greater than the number threshold, or the number of non-real objects is greater than the number threshold, or the number of non-real objects is greater than the number threshold, and the ratio of the number of non-real objects to the number of the at least one object Greater than the ratio threshold. Using this processing method can further improve the accuracy of object recognition.

In another example, the method provided in the embodiment of the present application may further include the following steps: for the first non-real object determined according to the first reflection intensity data set, according to historical object identification data and/or non-laser reflection intensity data To determine whether the first non-real object is a second non-real object.

For example, if a vehicle is determined to be a real vehicle based on the object recognition data and location data of the first 10 frames, but the vehicle is determined to be a water fog object based on the data of the current frame, the vehicle can be determined based on the object recognition data of the first 10 frames. The water mist object is a real object. Or, if a certain water-fog object is determined based on the object recognition data of the first 10 frames, and the object is still determined to be a water-fog object based on the data of the current frame, it can be determined that the water-fog object is indeed water based on the object recognition data of the first 10 frames Fog objects.

In the method provided by the embodiment of the present application, through the above step of judging whether the first non-real object is the second non-real object, if it is determined that the first non-real object determined in step S105 is a real object, the non-real object can be effectively improved. The safety of human driving.

It can be seen from the foregoing embodiments that the object detection method provided by the embodiments of the present application determines at least one object according to at least three-dimensional coordinate data in the road environment point cloud data; obtains the first reflection intensity data set of the object; The first reflection intensity data set is used to determine whether the object is a real object; this processing method makes effective use of the reflection intensity value information of the lidar, and performs false detection of the target based on the statistical information of the reflection intensity value of the target object , To avoid relying on other sensor data, but directly process the lidar point cloud to achieve a full range of unreal object filtering; therefore, it can effectively improve the accuracy of object detection.

Second embodiment

In the foregoing embodiment, an object detection method is provided. Correspondingly, the present application also provides an object detection device. This device corresponds to the embodiment of the above method. Since the device embodiment is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment. The device embodiments described below are merely illustrative.

This application additionally provides an object detection device, including:

The third embodiment

In the above-mentioned embodiment, an object detection method is provided. Correspondingly, this application also provides an object detection device. Since the device embodiment is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment. The device embodiments described below are merely illustrative.

An object detection device of this embodiment includes: a processor and a memory; the memory is used to store a program for implementing an object detection method. After the device is powered on and runs the program of the method through the processor, it executes The following steps: determine at least one object based on at least the three-dimensional coordinate data in the road environment point cloud data; acquire a first reflection intensity data set of the object; determine whether the object is based on the first reflection intensity data set Real objects.

The object detection device may be an unmanned vehicle, or a road test sensing device or the like.

Fourth embodiment

In the above-mentioned embodiment, an object detection method is provided. Correspondingly, the present application also provides an object detection method.

Please refer to FIG. 12, which is a flowchart of an embodiment of an object detection method provided by this application. The execution subject of the method includes but is not limited to an unmanned vehicle, and may also be a road test sensing device and other devices. An object detection method provided by this application includes:

Step S1201: Determine at least one object according to the road environment point cloud data.

Step S1203: divide the object into multiple sub-object blocks.

Step S1205: Determine whether the sub-object block is a real object block according to the reflection intensity data set of the sub-object block.

In specific implementation, step S1205 may include the following sub-steps: 1) According to the reflection intensity data set, determine the ratio of the number of point clouds respectively corresponding to a plurality of preset reflection intensity values to form the ratio of the number of point clouds of the sub-object block Vector; 2) through the object reality category prediction model, the reality category of the sub-object block is determined according to the point cloud quantity ratio vector.

Step S1207: Determine the second point cloud data of the object according to the first point cloud data of the object and the above judgment result.

In specific implementation, step S1207 may adopt the following manner: if the sub-object block is an unreal object, the point cloud data of the sub-object block is cleared from the first point cloud data of the object.

The above-mentioned steps of this embodiment are similar to the part of the first mode in the first embodiment, and will not be repeated here.

Fifth embodiment

This application additionally provides an object detection device, including:

Sixth embodiment

An object detection device of this embodiment includes: a processor and a memory; the memory is used to store a program for implementing an object detection method. After the device is powered on and runs the program of the method through the processor, it executes The following steps: determine at least one object based on the road environment point cloud data; divide the object into multiple sub-object blocks; determine whether the sub-object block is a real object block according to the reflection intensity data set of the sub-object block ; According to the first point cloud data of the object and the above judgment result, the second point cloud data of the object is determined.

Seventh embodiment

In the above-mentioned embodiment, an object detection method is provided. Correspondingly, this application also provides an object detection system.

Please refer to FIG. 13, which is a schematic diagram of an embodiment of an object detection system provided by this application. The system includes a terminal device 131 and a server 132.

The terminal device may be an unmanned vehicle, or a road test sensing device or the like.

The terminal device is used to collect road environment point cloud data, send an object detection request for the road environment point cloud data to the server; and receive real object point cloud data information sent back by the server; accordingly, the service The terminal is used to receive the request, determine at least one object based on at least the three-dimensional coordinate data in the road environment point cloud data; obtain the reflection intensity data set of the object; determine the object according to the reflection intensity data set Whether it is a real object; sending back real object point cloud data information to the terminal device.

Eighth embodiment

In the foregoing embodiment, an object detection system is provided. Correspondingly, the present application also provides an object detection method. The execution subject of this method includes but is not limited to unmanned vehicles, and can also be other devices such as road test sensing devices. An object detection method provided by this application includes: 1) collecting road environment point cloud data; 2) sending an object detection request for the road environment point cloud data to a server; 3) receiving a real object sent back by the server Point cloud data information.

Ninth embodiment

In the foregoing embodiment, an object detection method is provided. Correspondingly, the present application also provides an object detection device. An object detection device provided by this application includes:

Data collection unit, used to collect road environment point cloud data;

Tenth embodiment

In the above-mentioned embodiment, an object detection method is provided. Correspondingly, this application also provides an object detection device. An object detection device provided by this application includes:

Processor; and

Eleventh embodiment

In the foregoing embodiment, an object detection system is provided. Correspondingly, the present application also provides an object detection method. The execution subject of the method includes but is not limited to the server, and may also be other devices capable of implementing the method. An object detection method provided by this application includes: 1) receiving an object detection request for road environment point cloud data; 2) determining at least one object at least according to the three-dimensional coordinate data in the road environment point cloud data; 3) obtaining The reflection intensity data set of the object; 4) judging whether the object is a real object according to the reflection intensity data set; 5) sending back real object point cloud data information to the requesting party.

Twelfth embodiment

Thirteenth embodiment

Processor; and

Fourteenth embodiment

Please refer to FIG. 14, which is a schematic diagram of an embodiment of an object detection system provided by this application. The system includes a terminal device 141 and a server 142.

The terminal device is used to collect road environment point cloud data, send an object detection request for the road environment point cloud data to the server; and receive object point cloud data returned by the server; correspondingly, the server uses Upon receiving the request, determine at least one object according to the road environment point cloud data; divide the object into a plurality of sub-object blocks; obtain the reflection intensity data set of the sub-object blocks, and determine whether the sub-object blocks are It is a real object block; the second point cloud data of the object is determined according to the first point cloud data of the object and the above judgment result; the second point cloud data is returned to the terminal device.

Fifteenth embodiment

In the foregoing embodiment, an object detection system is provided. Correspondingly, the present application also provides an object detection method. The execution subject of this method includes but is not limited to unmanned vehicles, and can also be other devices such as road test sensing devices. An object detection method provided in this application includes: 1) collecting road environment point cloud data; 2) sending an object detection request for the road environment point cloud data to a server; 3) receiving object points sent back by the server Cloud data information.

Sixteenth embodiment

Data collection unit, used to collect road environment point cloud data;

Seventeenth embodiment

Processor; and

Eighteenth embodiment

In the foregoing embodiment, an object detection system is provided. Correspondingly, the present application also provides an object detection method. The execution subject of the method includes but is not limited to the server, and may also be other devices capable of implementing the method. An object detection method provided by this application includes: 1) receiving an object detection request for road environment point cloud data; 2) determining at least one object according to the road environment point cloud data; 3) dividing the object into multiple sub-objects Block; 4) determine whether the sub-object block is a real object block according to the reflection intensity data set of the sub-object block; 5) determine the object's value according to the first point cloud data of the object and the above judgment result The second point cloud data; 6) The second point cloud data is sent back to the requesting party.

Nineteenth embodiment

Twentieth embodiment

Processor; and

Twenty-first embodiment

In the foregoing embodiment, an object detection method is provided. Correspondingly, this application also provides a light source adjustment method. Since the method embodiment is basically similar to the method embodiment 1, the description is relatively simple, and the relevant part can refer to the part of the description of the method embodiment 1. The method embodiments described below are only illustrative.

Please refer to FIG. 15, which is a flowchart of an embodiment of a light source adjustment method provided by this application. The execution subject of the method includes, but is not limited to, unmanned vehicles, and may also be other devices such as road test sensing devices. A light source adjustment method provided by this application includes:

Step S1501: Determine at least one object according to the road environment data.

The road environment data includes, but is not limited to: road environment point cloud data collected by a laser radar or a camera scanner, road environment images collected by a camera, and so on.

To determine at least one object based on road environment data, a more mature existing technology can be used, for example, the RefineDet method based on deep learning can be used to identify at least one object from the road environment point cloud data; or The object detection model of the convolutional network recognizes at least one object from the road environment image.

Step S1503: Determine the number of non-real objects in the at least one object.

This step may be to perform further object authenticity recognition on at least one object determined in the previous step. For example, according to the first embodiment of the above method, it is judged whether the object is a real object according to the laser reflection intensity data of the object, or it is judged by a traditional method. Whether an object is a real object, after judging the authenticity of each object, the number of non-real objects in at least one object can be counted.

Step S1505: Adjust the working mode of the incident light source of the object according to the number of the unreal objects.

The number of non-real objects can reflect the road environment to a certain extent. Generally, the more the number of non-real objects, the worse the road environment (such as large water fog, severe dust, etc.), and the brightness of the road around the vehicle needs to be improved. , To improve the road environment, so that higher-quality perception data can be collected by sensors (such as cameras, etc.), and then the authenticity of the object can be recognized based on the higher-quality perception data, thereby improving the accuracy of the object authenticity judgment result And improve the safety of unmanned driving.

In this embodiment, this step can be implemented in the following manner: if the number of non-real objects meets the light source adjustment condition, the working mode of the incident light source of the object is adjusted.

The light source adjustment condition may be that the number of unreal objects is greater than the number threshold, or the ratio of the number of unreal objects to the number of the at least one object is greater than the ratio threshold (such as 20%), or the number of unreal objects is greater than the number threshold, And the ratio of the number of unreal objects to the number of the at least one object is greater than the ratio threshold.

In an example, at least one of the following methods may be used for adjusting the working mode of the incident light source of the object:

1) Turn on or turn off the lights of the first vehicle.

The first vehicle includes a vehicle that performs the method. For example, if the number of unreal objects is greater than the first number threshold (such as 3), turn on the lights of the first vehicle, which can enhance the light brightness of the road environment; if the number of unreal objects is less than or equal to the second number threshold (such as 1), Turn off the lights of the first vehicle, which can save vehicle power.

2) Adjust the brightness of the lights of the first vehicle.

For example, if the number of unreal objects is greater than the third number threshold (such as 5), increase the brightness of the lights of the first vehicle, which can effectively enhance the light brightness of the road environment; if the number of unreal objects is less than or equal to the fourth number threshold (such as 2), then reduce the brightness of the lights of the first vehicle, which can save vehicle power.

In another example, the adjusting the working mode of the incident light source of the object may include the following steps:

1) Determine the target second vehicle.

There may be multiple second vehicles around the first vehicle, from which the target second vehicle that can contribute to the improvement of the road environment of the first vehicle can be determined. The target second vehicle includes, but is not limited to, vehicles adjacent to the first vehicle, such as vehicles adjacent to the front, vehicles adjacent to left and right lanes, and so on.

2) Send the first instruction information for turning on or off the lights of the target second vehicle to the target second vehicle through the signal sending device, so that the target second vehicle receives the first instruction information through the signal receiving device, and according to the first instruction information The instruction information turns on or off the lights of the target second vehicle.

The first vehicle includes a signal sending device, and the second vehicle includes a signal receiving device, such as a wireless communication module. The first vehicle may send first indication information for turning on or off the lights of the target second vehicle to the target second vehicle through the signal sending device. The first indication information may include a field for indicating turning on or off the lights, and the target The second vehicle receives the first instruction information through the signal receiving device, and parses it to obtain the content of the field for instructing turning on or turning off the lights, thereby determining whether to turn on or off the lights of the target second vehicle.

For example, if the first vehicle determines to turn on the rear taillights of the second vehicle ahead based on the number of unreal objects, the first instruction message sent to the second vehicle includes the information for turning on the rear taillights, which can effectively improve the front of the first vehicle. The light intensity of the road environment.

For another example, if the first vehicle determines to turn off the rear taillights of the second vehicle ahead according to the number of unreal objects, the first instruction message sent to the second vehicle includes the information to turn off the rear taillights, so as to ensure that the front of the first vehicle In the case of the light intensity of the road environment, the power of the second vehicle can be effectively saved.

In yet another example, the adjusting the working mode of the incident light source of the object may include the following steps: 1) determining a target second vehicle, and the target second vehicle includes a vehicle adjacent to the first vehicle; 2) The signal sending device sends the second instruction information for adjusting the brightness of the lamp of the target second vehicle to the target second vehicle, so that the target second vehicle receives the second instruction information through the signal receiving device, and adjusts according to the second instruction information Target the brightness of the lights of the second vehicle.

For example, if the first vehicle determines to enhance the brightness of the rear taillights of the second vehicle ahead based on the number of unreal objects, the second indication information sent to the second vehicle includes information for enhancing the brightness of the rear taillights, which can effectively improve the first The light intensity of the road environment in front of the vehicle.

For another example, if the first vehicle determines to reduce the brightness of the rear taillights of the second vehicle ahead according to the number of unreal objects, the second indication message sent to the second vehicle includes information to reduce the brightness of the rear taillights, so as to ensure the first In the case of the light intensity of the road environment in front of the vehicle, the power of the second vehicle can be effectively saved.

In yet another example, the adjusting the working mode of the incident light source of the object may include the following steps: 1) determining a target street light, the target street light including a vehicle adjacent to the first vehicle; 2) using a signal sending device The third instruction information for turning on or off the target street light is sent to the target street light, so that the target street light receives the third instruction information through the signal receiving device, and turns on or off the target street light according to the third instruction information.

For example, if the first vehicle determines to turn on the nearest street light based on the number of unreal objects, the third instruction information sent to the nearest street light includes information on turning on the street light, which can effectively improve the light intensity of the road environment in front of the first vehicle.

For another example, if the first vehicle determines to turn off the nearest street light based on the number of unreal objects, the third instruction message sent to the nearest street light includes the information of turning off the street light, so as to ensure the light intensity of the road environment in front of the first vehicle This can effectively save the energy of the nearest street lamp (such as electricity consumption, etc.).

During specific implementation, the target street lamp can be determined according to the position information of the street lamp and the position information of the first vehicle. The location information of the street lights can be obtained through the map server.

The street light includes a signal receiving device, a processor, and a memory, and the processor performs the following steps: receiving the third instruction information through the signal receiving device, and turning on or off the street light according to the third instruction information.

In another example, the adjusting the working mode of the incident light source of the object may include the following steps: 1) determining a target street light, the target street light including a vehicle adjacent to the first vehicle; 2) using a signal sending device The fourth instruction information for adjusting the brightness of the street light is sent to the target street light, so that the target street light receives the fourth instruction information through the signal receiving device, and adjusts the brightness of the street light according to the fourth instruction information.

For example, if the first vehicle determines to enhance the brightness of the nearest street lamp based on the number of unreal objects, the fourth instruction information sent to the nearest street lamp includes information to enhance the brightness of the street lamp, which can effectively improve the light of the road environment in front of the first vehicle strength.

For another example, if the first vehicle determines to reduce the brightness of the nearest street lamp based on the number of unreal objects, the fourth instruction information sent to the nearest street lamp includes information to reduce the brightness of the street lamp, so as to ensure the light of the road environment in front of the first vehicle. In the case of high intensity, it can effectively save the power of the nearest street lamp.

It can be seen from the above embodiment that the light source adjustment method provided by the embodiment of the present application determines at least one object according to road environment data; determines the number of non-real objects in the at least one object; and adjusts all the objects according to the number of non-real objects. The working mode of the incident light source of the object; this processing mode can improve the environment of the vehicle driving road in real time, so as to collect higher-quality road environment perception data; therefore, the safety of unmanned driving can be effectively improved.

Twenty-second embodiment

In the above-mentioned embodiment, a light source adjustment method is provided. Correspondingly, the present application also provides a light source adjustment device. Since the device embodiment is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment. The device embodiments described below are merely illustrative. A light source adjustment device provided by this application includes:

Twenty-third embodiment

In the above-mentioned embodiment, a light source adjustment method is provided. Correspondingly, the present application also provides a vehicle. Since the device embodiment is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment. The device embodiments described below are merely illustrative.

A vehicle of this embodiment includes: a processor and a memory; the memory is used to store a program for realizing the light source adjustment method. After the device is powered on and the program of the method is run through the processor, the following steps are executed: Determine at least one object according to the road environment data; determine the number of non-real objects in the at least one object; and adjust the working mode of the incident light source of the object according to the number of non-real objects.

Twenty-fourth embodiment

In the above-mentioned embodiment, a light source adjustment method is provided. Correspondingly, this application also provides a light source adjustment method. Since this method embodiment is basically similar to the above method embodiment, the description is relatively simple, and for related parts, please refer to the partial description of the above method embodiment. The method embodiments described below are only illustrative. A light source adjustment method provided by the present application includes: 1) receiving light source adjustment instruction information; 2) adjusting the working mode of the light source attached to the vehicle according to the instruction information.

Twenty-fifth embodiment

A vehicle of this embodiment includes: a signal receiving device, a processor, and a memory; the memory is used to store a program that implements the light source adjustment method. After the device is powered on and the program of the method is run through the processor, the program is executed The following steps: receiving light source adjustment instruction information; according to the instruction information, adjusting the working mode of the light source attached to the vehicle.

Twenty-sixth embodiment

In the above-mentioned embodiment, a light source adjustment method is provided. Correspondingly, the present application also provides a light source adjustment method. The execution subject of the method includes a street lamp. Since this method embodiment is basically similar to the above method embodiment, the description is relatively simple, and for related parts, please refer to the partial description of the above method embodiment. The method embodiments described below are only illustrative. A light source adjustment method provided by the present application includes: 1) receiving light source adjustment instruction information; 2) adjusting the working mode of the street light according to the instruction information.

Twenty-seventh embodiment

In the above-mentioned embodiment, a light source adjustment method is provided. Correspondingly, this application also provides a street lamp. Since the device embodiment is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment. The device embodiments described below are merely illustrative.

A street lamp of this embodiment includes: a signal receiving device, a processor, and a memory; the memory is used to store a program that implements the light source adjustment method. After the device is powered on and the method is run through the processor, the program is executed The following steps: receiving light source adjustment instruction information; adjusting the operating mode of the street lamp according to the instruction information.

Twenty-eighth embodiment

In the above-mentioned embodiment, an object detection method is provided. Correspondingly, the present application also provides an object detection method. Since the method embodiment is basically similar to the method embodiment 1, the description is relatively simple, and the relevant part can refer to the part of the description of the method embodiment 1. The method embodiments described below are only illustrative.

Please refer to FIG. 16, which is a flowchart of an embodiment of an object detection method provided by this application. The execution subject of the method includes but is not limited to an unmanned vehicle, and may also be a road test sensing device and other devices. An object detection method provided by this application includes:

Step S1601: Determine at least one object according to the road environment data.

The road environment data includes, but is not limited to: road environment point cloud data collected by lidar or camera scanner, road environment images collected by cameras, sound wave data collected by ultrasonic sensors, and so on.

Step S1603: Acquire sound wave reflection intensity data of the object.

The sound wave reflection intensity data may be an echo that is reflected by the ultrasonic wave emitted from an ultrasonic sensor and after the ultrasonic wave hits the surface of an object around the vehicle. Ultrasound has a great ability to penetrate liquids (water mist) and dust, but it will produce significant reflections after encountering objects such as vehicles and pedestrians to form reflected echoes, and moving objects can produce Doppler effects.

The sound wave reflection intensity data may also be the sound played through a speaker, and the sound wave reflected back after the sound reaches the surface of the object.

In this embodiment, step S1603 may include the following steps: 1) collecting sound wave reflection intensity data of the road environment through an ultrasonic sensor; 2) determining sound source position information according to the sound wave reflection intensity data; 3) according to the sound wave reflection intensity data The source position information and the three-dimensional coordinate data of the object determine the sound wave reflection intensity data of the object.

The sound source location information can be determined according to the sound wave reflection intensity data using relatively mature existing technology, which will not be repeated here.

Step S1605: Determine whether the object is a real object according to the sound wave reflection intensity data.

For example, if the distance between the sound source location information and the three-dimensional coordinate data of the object is less than the distance threshold, it means that the reflected sound wave originates from the object. Therefore, the object can be determined to be a real object and be regarded as a real object; if If the distance is greater than or equal to the distance threshold, it means that the sound wave does not originate from the object, and the ultrasonic wave emitted by the ultrasonic sensor does not reflect back, but penetrates the object. Therefore, it can be determined that the object is not a real object.

It can be seen from the above-mentioned embodiments that the method provided by the embodiments of this application determines at least one object based on road environment data; obtains the sound wave reflection intensity data of the object; and determines whether the object is real according to the sound wave reflection intensity data Object; this processing method makes effective use of the sound wave reflection intensity value information, and according to the sound wave reflection intensity data of the target object to filter out the false detection target, to achieve a full range of non-real object filtering; therefore, it can effectively improve the object The accuracy of detection.

Twenty-ninth embodiment

In the foregoing embodiment, an object detection method is provided. Correspondingly, the present application also provides an object detection device. Since the device embodiment is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment. The device embodiments described below are merely illustrative. An object detection device provided by this application includes:

Thirtieth embodiment

An object detection device of this embodiment includes: a processor and a memory; the memory is used to store a program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, it executes the following Steps: determine at least one object based on at least three-dimensional coordinate data in the road environment point cloud data; obtain a first reflection intensity data set of the object; determine whether the object is a real object according to the first reflection intensity data set .

Although this application is disclosed as above in preferred embodiments, it is not intended to limit the application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the application. Therefore, this application The scope of protection shall be subject to the scope defined by the claims of this application.

In a typical configuration, the computing device includes one or more processors (CPU), input/output interfaces, network interfaces, and memory.

The memory may include non-permanent memory in computer readable media, random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of computer readable media.

1. Computer-readable media include permanent and non-permanent, removable and non-removable media, and information storage can be realized by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, CD-ROM, digital versatile disc (DVD) or other optical storage, Magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media can be used to store information that can be accessed by computing devices. According to the definition in this article, computer-readable media does not include non-transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

2. Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems or computer program products. Therefore, this application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

Claims

An object detection method, characterized by comprising:

Determine at least one object at least according to the three-dimensional coordinate data in the road environment point cloud data;

Acquiring a first reflection intensity data set of the object;

According to the first reflection intensity data set, it is determined whether the object is a real object.
The method of claim 1, wherein the judging whether the object is a real object according to the first reflection intensity data set comprises:

Determine, according to the first reflection intensity data set, a first point cloud quantity ratio corresponding to a plurality of preset reflection intensity values, to form a first point cloud quantity ratio vector of the object;

The object authenticity category prediction model determines the authenticity category of the object according to the first point cloud quantity ratio vector; the prediction model labels data from the first point cloud quantity ratio vector of the multiple objects and the object authenticity category Learned from the correspondence between.
The method according to claim 2, further comprising:

Based on the weighted cross-entropy loss function, the prediction model is learned from a plurality of the corresponding relationships; wherein, the first loss weight corresponding to the loss caused by misjudging a real object as an unreal object is greater than that of misjudging an unreal object It is judged as the second loss weight corresponding to the loss caused by the real object.
The method according to claim 2, further comprising:

Dividing the object into multiple sub-object blocks;

Acquiring a second reflection intensity data set of the sub-object block;

Determine, according to the second reflection intensity data set, a second point cloud quantity ratio corresponding to the plurality of preset reflection intensity values, to form a second point cloud quantity ratio vector of the sub-object block;

Using the prediction model to determine the authenticity category of the sub-object block according to the second point cloud quantity ratio vector;

According to the first point cloud data of the object and the authenticity category of each sub-object block, the second point cloud data of the object is determined.
The method of claim 1, wherein the judging whether the object is a real object according to the first reflection intensity data set comprises:

Determine, according to the first reflection intensity data set, the number of first point clouds whose reflection intensity data of the object is greater than a threshold value of real object reflection intensity data;

According to the number of the first point cloud, it is determined whether the object is a real object.
The method of claim 5, wherein the judging whether the object is a real object according to the number of the first point cloud comprises:

Acquiring a third point cloud quantity ratio between the first point cloud quantity and the second point cloud quantity of the object;

If the third point cloud quantity ratio is greater than the real object ratio threshold, the object is regarded as a real object.
The method according to claim 6, further comprising:

If the third point cloud quantity ratio is less than the ratio threshold, determining the reflection intensity value quantity of the non-zero point cloud quantity of the object;

If the number of reflection intensity values is greater than the threshold value of the number of reflection intensity values, the object is regarded as a real object.
The method of claim 1, wherein the judging whether the object is a real object according to the first reflection intensity data set comprises:

For each object, determine the number of reflection intensity values of the number of non-zero point clouds of the object;

If the number of reflection intensity values is greater than the threshold value of the number of reflection intensity values, the object is regarded as a real object.
The method according to claim 1, wherein before said acquiring the first reflection intensity data set of the object, the method further comprises:

Determining an object that affects the driving mode of the vehicle from the at least one object;

The judging whether the object is a real object according to the first reflection intensity data set includes:

For the object that affects the driving mode of the vehicle, it is determined whether the object is a real object according to the first reflection intensity data set.
The method of claim 9, wherein the object that affects the driving mode of the vehicle comprises:

Objects located in front of the vehicle, objects located in the adjacent lane of the vehicle.
The method according to claim 1, further comprising:

If the number of unreal objects determined according to the first reflection intensity data set satisfies the alarm condition, the alarm information is displayed, so that the user controls the vehicle to enter the non-unmanned driving mode according to the alarm information.
The method according to claim 1, further comprising:

Regarding the first real object determined according to the first reflection intensity data set, it is determined whether the first real object is a second real object according to historical object recognition data and/or non-laser reflection intensity data.
The method according to claim 1, further comprising:

For the first non-real object determined according to the first reflection intensity data set, it is determined whether the first non-real object is a second non-real object according to historical object recognition data and/or non-laser reflection intensity data.
An object detection method, characterized by comprising:

Determine at least one object according to the road environment point cloud data;

Dividing the object into multiple sub-object blocks;

Judging whether the sub-object block is a real object block according to the reflection intensity data set of the sub-object block;

According to the first point cloud data of the object and the above judgment result, the second point cloud data of the object is determined.
The method according to claim 14, wherein the judging whether the sub-object block is a real object block according to the reflection intensity data set comprises:

According to the reflection intensity data set, determine the ratio of the number of point clouds respectively corresponding to a plurality of preset reflection intensity values, and form the ratio vector of the number of point clouds of the sub-object block;

Through the object authenticity category prediction model, the authenticity category of the sub-object block is determined according to the point cloud quantity ratio vector.
An object detection device, characterized by comprising:

The object prediction unit is configured to determine at least one object at least according to the three-dimensional coordinate data in the road environment point cloud data;

A data acquisition unit for acquiring a first reflection intensity data set of the object;

The real object determining unit is configured to determine whether the object is a real object according to the first reflection intensity data set.
An object detection device, characterized in that it comprises:

Processor; and

The memory is used to store the program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: at least according to the three-dimensional coordinate data in the road environment point cloud data, determine at least An object; obtain a first reflection intensity data set of the object; determine whether the object is a real object according to the first reflection intensity data set.
The device according to claim 17, wherein the device comprises: a vehicle or a drive test sensing device.
An object detection device, characterized by comprising:

The object prediction unit is used to determine at least one object according to the road environment point cloud data;

An object segmentation unit for segmenting the object into multiple sub-object blocks;

The real object block determining unit is configured to determine whether the sub object block is a real object block according to the reflection intensity data set of the sub object block;

The object determining unit is used for the object to determine the second point cloud data of the object according to the first point cloud data of the object and the above judgment result.
An object detection device, characterized in that it comprises:

Processor; and

The memory is used to store a program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: determine at least one object according to the road environment point cloud data; The object is divided into multiple sub-object blocks; according to the reflection intensity data set of the sub-object blocks, it is judged whether the sub-object blocks are real object blocks; according to the first point cloud data of the object and the above judgment result, the The second point cloud data of the object.
An object detection system is characterized in that it comprises:

A terminal device for collecting road environment point cloud data of the terminal device, sending an object detection request for the road environment point cloud data to the server; and receiving real object point cloud data information returned by the server;

The server is configured to receive the request, determine at least one object based on at least the three-dimensional coordinate data in the road environment point cloud data; obtain the reflection intensity data set of the object; determine the reflection intensity data set according to the reflection intensity data set. Whether the object is a real object; sending back the point cloud data information of the real object to the terminal device.
An object detection system is characterized in that it comprises:

The terminal device is configured to collect the road environment point cloud data of the terminal device, send an object detection request for the road environment point cloud data to the server; and receive the object point cloud data returned by the server;

The server is configured to receive the request, determine at least one object based on the road environment point cloud data; divide the object into multiple sub-object blocks; determine the sub-object based on the reflection intensity data set of the sub-object blocks Whether the block is a real object block; determine the second point cloud data of the object according to the first point cloud data of the object and the above judgment result; send the second point cloud data back to the terminal device.
An object detection method, characterized by comprising:

Collect road environment point cloud data;

Sending an object detection request for the road environment point cloud data to the server;

Receiving real object point cloud data information returned by the server.
An object detection method, characterized by comprising:

Receive object detection requests for road environment point cloud data;

At least determining at least one object according to the three-dimensional coordinate data in the road environment point cloud data;

Acquiring a reflection intensity data set of the object;

Judging whether the object is a real object according to the reflection intensity data set;

Send back the point cloud data information of the real object to the requesting party.
An object detection device, characterized by comprising:

Data collection unit, used to collect road environment point cloud data;

A request sending unit, configured to send an object detection request for the road environment point cloud data to the server;

The data receiving unit is configured to receive real object point cloud data information returned by the server.
An object detection device, characterized in that it comprises:

Processor; and

The memory is used to store the program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: collect road environment point cloud data; An object detection request for environmental point cloud data; receiving real object point cloud data information returned by the server.
An object detection device, characterized by comprising:

The request receiving unit is configured to receive an object detection request for road environment point cloud data;

An object prediction unit, configured to determine at least one object at least according to the three-dimensional coordinate data in the road environment point cloud data;

A data acquisition unit for acquiring the reflection intensity data set of the object;

A real object determining unit, configured to determine whether the object is a real object according to the reflection intensity data set;

The data return unit is used to return real object point cloud data information to the requesting party.
An object detection device, characterized in that it comprises:

Processor; and

The memory is used to store the program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: receiving an object detection request for road environment point cloud data; According to the three-dimensional coordinate data in the road environment point cloud data, at least one object is determined; the reflection intensity data set of the object is acquired; the reflection intensity data set is used to determine whether the object is a real object; the real object is sent back to the requesting party Point cloud data information.
An object detection method, characterized by comprising:

Collect road environment point cloud data;

Sending an object detection request for the road environment point cloud data to the server;

Receiving the object point cloud data information returned by the server.
An object detection method, characterized by comprising:

Receive object detection requests for road environment point cloud data;

Determine at least one object according to the road environment point cloud data;

Dividing the object into multiple sub-object blocks;

Judging whether the sub-object block is a real object block according to the reflection intensity data set of the sub-object block;

Determine the second point cloud data of the object according to the first point cloud data of the object and the foregoing judgment result;

Send the second point cloud data back to the requesting party.
An object detection device, characterized by comprising:

Data collection unit, used to collect road environment point cloud data;

A request sending unit, configured to send an object detection request for the road environment point cloud data to the server;

The data receiving unit is used to receive the object point cloud data information returned by the server.
An object detection device, characterized in that it comprises:

Processor; and

The memory is used to store the program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: collect road environment point cloud data; Object detection request for environmental point cloud data; receiving object point cloud data information returned by the server.
An object detection device, characterized by comprising:

The request receiving unit is configured to receive an object detection request for road environment point cloud data;

The object prediction unit is used to determine at least one object according to the road environment point cloud data;

An object segmentation unit for segmenting the object into multiple sub-object blocks;

The real object block determining unit is configured to determine whether the sub object block is a real object block according to the reflection intensity data set of the sub object block;

An object determining unit, configured to determine the second point cloud data of the object according to the first point cloud data of the object and the above judgment result;

The data return unit is configured to return the second point cloud data to the requesting party.
An object detection device, characterized in that it comprises:

Processor; and

The memory is used to store the program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: receiving an object detection request for the road environment point cloud data; according to the road environment Point cloud data, determine at least one object; divide the object into multiple sub-object blocks; determine whether the sub-object block is a real object block according to the reflection intensity data set of the sub-object block; The point cloud data and the above judgment result determine the second point cloud data of the object; the second point cloud data is sent back to the requesting party.
A method for adjusting a light source is characterized by comprising:

Determine at least one object based on road environment data;

Determining the number of non-real objects in the at least one object;

According to the number of the unreal objects, the working mode of the incident light source of the object is adjusted.
The method according to claim 35, wherein the adjusting the working mode of the incident light source of the object according to the number of the unreal objects comprises:

If the number of unreal objects meets the light source adjustment condition, the working mode of the incident light source of the object is adjusted.
The method of claim 36, wherein the light source adjustment condition comprises:

The number of non-real objects is greater than the number threshold, and/or the ratio between the number of non-real objects and the number of the at least one object is greater than the ratio threshold.
The method according to claim 35, wherein the adjusting the working mode of the incident light source of the object adopts at least one of the following modes:

Turning on or off the lights of a first vehicle, the first vehicle including a vehicle that executes the method;

Adjust the brightness of the lights of the first vehicle.
The method of claim 35, wherein the adjusting the working mode of the incident light source of the object comprises:

Determining a target second vehicle, where the target second vehicle includes a vehicle adjacent to the first vehicle;

The first instruction information for turning on or off the lights of the target second vehicle is sent to the target second vehicle through the signal sending device, so that the target second vehicle receives the first instruction information through the signal receiving device, and according to the first instruction information Turn on or off the lights of the target second vehicle.
The method of claim 35, wherein the adjusting the working mode of the incident light source of the object comprises:

Determining a target second vehicle, where the target second vehicle includes a vehicle adjacent to the first vehicle;

The signal sending device sends the second instruction information for adjusting the brightness of the lamp of the target second vehicle to the target second vehicle, so that the target second vehicle receives the second instruction information through the signal receiving device, and adjusts according to the second instruction information Target the brightness of the lights of the second vehicle.
The method of claim 35, wherein the adjusting the working mode of the incident light source of the object comprises:

Determining a target street light, where the target street light includes a vehicle adjacent to the first vehicle;

The signal sending device sends third instruction information for turning on or off the target street light to the target street light, so that the target street light receives the third instruction information through the signal receiving device, and turns on or off the target street light according to the third instruction information.
The method of claim 35, wherein the adjusting the working mode of the incident light source of the object comprises:

Determining a target street light, where the target street light includes a vehicle adjacent to the first vehicle;

The signal sending device sends the fourth instruction information for adjusting the brightness of the street light to the target street light, so that the target street light receives the fourth instruction information through the signal receiving device, and adjusts the brightness of the street light according to the fourth instruction information.
A light source adjusting device is characterized by comprising:

The object prediction unit is used to determine at least one object according to road environment data;

A unit for determining the number of non-real objects, configured to determine the number of non-real objects in the at least one object;

The light source adjusting unit is used to adjust the working mode of the incident light source of the object according to the number of the unreal object.
A vehicle, characterized by comprising:

Processor; and

The memory is used to store a program for realizing the light source adjustment method. After the vehicle is powered on and runs the light source adjustment method program through the processor, the following steps are executed: determine at least one object according to road environment data; determine the at least one The number of non-real objects in the object; according to the number of non-real objects, the working mode of the incident light source of the object is adjusted.
A vehicle, characterized by comprising:

Signal receiving device;

Processor; and

The memory is used to store a program for realizing the light source adjustment method. After the vehicle is powered on and runs the light source adjustment method program through the processor, the following steps are executed: receiving light source adjustment instruction information; adjusting the light source adjustment method according to the instruction information How the light source attached to the vehicle works.
A method for adjusting a light source is characterized by comprising:

Receive light source adjustment instruction information;

According to the instruction information, the working mode of the light source attached to the vehicle is adjusted.
A street lamp, characterized by comprising:

Signal receiving device;

Processor; and

The memory is used to store a program for realizing the light source adjustment method. After the street lamp is powered on and runs the light source adjustment method program through the processor, the following steps are executed: receiving light source adjustment instruction information; and adjusting the light source adjustment method according to the instruction information How street lights work.
A method for adjusting a light source is characterized by comprising:

Receive light source adjustment instruction information;

According to the instruction information, adjust the working mode of the street lamp.
An object detection method, characterized by comprising:

Determine at least one object based on road environment data;

Acquiring sound wave reflection intensity data of the object;

According to the sound wave reflection intensity data, it is determined whether the object is a real object.
The method according to claim 49, wherein said acquiring sound wave reflection intensity data of said object comprises:

Collect sound wave reflection intensity data of the road environment through ultrasonic sensors;

Determine the sound source location information according to the sound wave reflection intensity data of the road environment;

According to the sound source position information and the three-dimensional coordinate data of the object, the sound wave reflection intensity data of the object is determined.
An object detection device, characterized by comprising:

The object prediction unit is used to determine at least one object according to road environment data;

A data acquisition unit for acquiring sound wave reflection intensity data of the object;

The real object determining unit is configured to determine whether the object is a real object according to the sound wave reflection intensity data.
An object detection device, characterized in that it comprises:

Processor; and

The memory is used to store the program for implementing the object detection method. After the device is powered on and runs the object detection method program through the processor, the following steps are executed: at least according to the three-dimensional coordinate data in the road environment point cloud data, determine at least An object; obtain a first reflection intensity data set of the object; determine whether the object is a real object according to the first reflection intensity data set.