CN114067552A - Pedestrian crossing track tracking and predicting method based on roadside laser radar - Google Patents
Pedestrian crossing track tracking and predicting method based on roadside laser radar Download PDFInfo
- Publication number
- CN114067552A CN114067552A CN202111312367.3A CN202111312367A CN114067552A CN 114067552 A CN114067552 A CN 114067552A CN 202111312367 A CN202111312367 A CN 202111312367A CN 114067552 A CN114067552 A CN 114067552A
- Authority
- CN
- China
- Prior art keywords
- neural network
- pedestrian
- layer
- trajectory
- track
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/005—Traffic control systems for road vehicles including pedestrian guidance indicator
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0108—Measuring and analyzing of parameters relative to traffic conditions based on the source of data
- G08G1/0116—Measuring and analyzing of parameters relative to traffic conditions based on the source of data from roadside infrastructure, e.g. beacons
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
- G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/04—Detecting movement of traffic to be counted or controlled using optical or ultrasonic detectors
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/16—Anti-collision systems
- G08G1/166—Anti-collision systems for active traffic, e.g. moving vehicles, pedestrians, bikes
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Analytical Chemistry (AREA)
- Chemical & Material Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention relates to a method for tracking and predicting a pedestrian crossing track based on a roadside laser radar, which belongs to the field of traffic safety and comprises the following steps: (1) acquiring track data; (2) preprocessing the track data acquired in the step (1); (3) training an artificial neural network model through the track data preprocessed in the step (2); (4) and preprocessing the trajectory data to be predicted and inputting the trajectory data into the trained artificial neural network model to obtain a prediction result. The method is based on roadside lidar point cloud data, provides prediction of the intention of pedestrians crossing/not crossing roads at different positions, has higher prediction precision compared with the traditional neural network model, and reduces the influence caused by possible failure of a vehicle-mounted sensing system when the vehicle-mounted sensing system is blocked by other vehicles and roadside objects.
Description
Technical Field
The invention relates to a method for tracking and predicting a pedestrian crossing track based on a roadside laser radar, and belongs to the field of traffic safety.
Background
Road traffic safety is a difficult problem troubling the traffic field, and the development of computer technology brings a new solution for traffic safety.
The trajectory tracking and the crossing intention prediction of the pedestrian at the crossing are important for the safety of the crossing. Vehicle mounted video sensors have been developed to detect pedestrians. However, vehicle-mounted systems cannot remind pedestrians to take evasive action when they are in danger. In recent years, advanced driving assistance systems have been explored and developed to improve driver and pedestrian safety. Vehicle mounted sensors have some inherent limitations such as the lack of a pedestrian warning mechanism to alert pedestrians to take evasive action in case of danger. In addition, on-board sensing systems may fail when blocked by other vehicles and roadside objects.
To extract useful information from image data, many models and computational algorithms have been developed to predict whether a pedestrian will cross a road. However, this complex task requires the detection, classification and tracking of the target object from a large number of candidate image regions, which is computationally expensive, especially for 3D video data processing. Vehicle-mounted systems can meet the requirements for high performance computing, while roadside systems are difficult to meet. In addition, there is no guarantee that the vehicle will be able to view each pedestrian and each obstacle under different lighting conditions. The performance of the camera is affected by lighting conditions, which means that the video quality may be lower at night than during the day.
Installing a lidar at the roadside is an innovative approach to solving the above-mentioned problems. The lidar sensor detects the position of an object by actively emitting a laser beam and measuring the propagation time after reflection. The laser radar sensor can stably and reliably work under different environmental conditions in the day and at night. The three-dimensional laser radar sensor can perform high-precision 360-degree three-dimensional scanning on surrounding objects. Lidar sensors have the advantages of reliability, accuracy, wide coverage and greatly reduced price, and have become a promising component of traffic infrastructure, particularly in autonomous driving and networked vehicle applications. The invention relates to a method for tracking and predicting a pedestrian crossing track based on a roadside laser radar.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a method for tracking and predicting the pedestrian crossing track based on the roadside laser radar, which is used for predicting the intention of pedestrians crossing/not crossing roads at different positions on the basis of roadside laser radar point cloud data. The trained model can be used for predicting the crossing intention of the pedestrian and providing timely alarm for the pedestrian and the vehicle through a roadside or vehicle-mounted early warning system.
Interpretation of terms:
background filtering, point cloud images scanned by the laser radar contain road surrounding environments (such as trees, signal lamps, building facilities and the like) and on-road targets, and in order to independently research the motion rules of the on-road targets, an algorithm adopted for filtering the point cloud data of the surrounding environments is needed, which is a background filtering method based on road traffic point cloud data from Chinese patent document CN 113447953A.
The method comprises the steps of object classification, construction and training of an artificial neural network for vehicle and pedestrian classification, taking the total number of point clouds, the distance from a laser radar and the direction of a target cluster formed by point clouds as input, excluding pedestrians from each picture through processing of an input layer, a hidden layer and an output layer, and only keeping the target cluster formed by point clouds of vehicles, and is a lane change identification and prediction method, a lane change identification and prediction system, lane change identification and prediction equipment and a storage medium for extracting vehicle tracks by using roadside laser radar data, which are disclosed in Chinese patent document CN 113345237A.
Lane recognition, generating lane boundary lines, and matching vehicles in transit into corresponding lanes; the Chinese patent document CN113345237A discloses a lane change recognition and prediction method, a system, equipment and a storage medium for extracting vehicle tracks by using roadside lidar data.
Road user clustering: the method, the system, the equipment and the storage medium for identifying and predicting lane change by using roadside laser radar data to extract vehicle tracks are disclosed in Chinese patent document CN 113345237A.
The DBSCAN algorithm is a spatial clustering algorithm based on density. The algorithm divides the area with sufficient density into clusters and finds arbitrarily shaped clusters in a spatial database with noise, which defines clusters as the largest set of density-connected points. It should be noted that the algorithm takes into account the spatial point cloud distribution, including the length, width and height of the clusters.
Track tracing: due to the linear propagation of the laser, the vehicle in a far lane is shielded by the vehicle in a near lane, so that the point cloud picture of the vehicle disappears or deforms in the three-dimensional point cloud. Therefore, in order to prevent repeated false recognition or missed recognition of the target, vehicles are tracked by a global distance search method, and a certain vehicle in the current frame is associated with the same vehicle in the previous frame, which is an in-transit target classification method based on the roadside laser radar in chinese patent document CN 113191459A.
A depth automatic encoder is an input data set low-dimensional representation method based on deep learning. The extracted features may more efficiently and abstractly represent features of the data. Therefore, the deep automatic encoder can be regarded as a pre-training process of the neural network model.
The technical scheme of the invention is as follows:
a pedestrian crossing track tracking and predicting method based on a roadside laser radar comprises the following operation steps:
(1) acquiring track data;
(2) preprocessing the track data acquired in the step (1);
(3) training an artificial neural network model through the track data preprocessed in the step (2);
(4) and preprocessing the trajectory data to be predicted and inputting the trajectory data into the trained artificial neural network model to obtain a prediction result.
Preferably, in step (1), historical trajectory data is extracted from pedestrian trajectory data of the laser radar, and data stream is converted into real-time trajectory of road users through background filtering, lane recognition, road user clustering, object classification and tracking, wherein the real-time trajectory comprises XYZ coordinate position (coordinates in data space of the laser radar), speed and direction information.
Preferably, in step (2), the road area (lane) and the sidewalk area are divided by using a linear boundary, as shown in fig. 1, and a point C on the boundary represents a starting point of the crosswalk. Along the sidewalk, the area is divided into rectangular areas with the same width, and the vertical distance from the pedestrian track point to the side of the sidewalk in the rectangular area is the distance from the pedestrian to the pedestrian crossing along the sidewalk direction.
It is further preferred that not all crossing points occur on the crosswalk and that some pedestrians choose to cross the road before reaching the crosswalk. The rectangular area is to be able to cover all pedestrians at different positions of the intersection.
Further preferably, the label determination criterion is that the distance between the trace point of the pedestrian at the last position of each region and the boundary of the sidewalk is determined, and if the distance is less than 0.3m, the label is 1, the pedestrian is considered to be crossing, and if the trace point is displayed on the side of the crosswalk of the road, the pedestrian is indicated to enter the road, and prediction is not needed, and the label is 1.
The distance threshold of 0.3m is determined by the average speed of the pedestrian and the rotation frequency of the lidar sensor (or the duration of one frame of data). A larger distance threshold may result in a false classification of a pedestrian not crossing the road as a pedestrian crossing the road, and a smaller distance threshold may result in a false classification of a pedestrian not crossing the road as a pedestrian crossing the road.
Preferably, in the step (3), the six characteristics of the XYZ position, the speed, the direction and the distance from the track point to the pedestrian crossing are used as the input of a depth self-encoder, the input is processed by a depth self-encoder to obtain a code, then the code and the label information are input into an artificial neural network, a prediction model is trained, and the trained prediction model is used for performing label distribution on the track to be predicted.
Preferably, in the step (4), the specific steps of predicting the trajectory by the artificial neural network are as follows:
4.1: constructing an artificial neural network for labeling the track;
4.2: making a data set, wherein the data set is divided into a training set, a verification set and a test set;
4.3: the artificial neural network for labeling the trajectory is trained through a training set, and the artificial neural network for labeling the trajectory is verified through a verification set. The verification process comprises the following steps: and (3) verifying the trajectory data in the verification set by using the artificial neural network trained by the training set, and if the verification effect is not ideal, continuing training by using the training set until the requirements are met. The trained artificial neural network for labeling the track can be obtained through the steps;
4.4: and inputting the test set into a trained artificial neural network for predicting the track to obtain a classification result, wherein the classification result comprises a road passing and a road not passing.
Further preferably, the artificial neural network for labeling the trajectory comprises an input layer, a hidden layer and an output layer;
the input layer is used for inputting the point cloud picture into an artificial neural network for labeling the track; the hidden layer comprises a plurality of layers of neural networks, convolution operation is carried out on the input point cloud image, point cloud information characteristics (the point cloud information characteristics of pedestrians comprise the point cloud number, the length, the width, the height and the density of a point cloud cluster) are extracted, and final overall characteristics (the shallow layer characteristics are a plurality of sections of lines formed by appearance contour points of the point cloud cluster, and the overall characteristics are the overall contours formed by combining the sections) are obtained through multiple times of convolution in the hidden layer, so that the tracks are labeled;
the output layer comprises two neurons, namely a label 1 and a label 0; the output layer is connected with the last layer of neural network of the hidden layer by adopting a full connection method, the probability of whether the pedestrian passes the road or not is obtained through parameter operation between the neurons, and the higher probability is used as a classification result.
It is further preferable that, in order to avoid the overfitting phenomenon of the network, a random factor is added to the neurons in each layer of the neural network, and in a certain layer of the neural network, the random factor randomly inhibits certain neurons in the layer of the neural network.
The process of adding random factor inhibition is as follows: in a certain layer of neurons, a plurality of neurons are randomly extracted, and the connection between the neurons and other neurons is cut off, namely, the neurons in the previous layer skip the neurons and are directly connected with the neurons in the next layer. For example, if there are 100 neurons in a layer and the random factor is 0.05, then 100 × 0.05 — 5 neurons are selected in that layer. The selected 5 neurons are inhibited, and the upper layer and the lower layer jump over the neurons to be directly connected. Random factors are introduced between each layer of neurons.
Even if it does not function, the neurons of the upper or lower layer connected to the neuron are not affected. This is because in the hidden layer, neurons in each layer are all connected, and when a neuron is inhibited, neurons in adjacent layers are not affected.
Further preferably, step 4.4 further includes comparing the actual output with the estimated output (actual behavior of the target to be detected), and the specific implementation process is as follows: and feeding back the error between the actual output and the estimated output to the input layer again to serve as a guide for adjusting the weight in the next round of training, adjusting the weight of each neuron in each layer of neural network through the iterative process, gradually learning the intrinsic relation between the input and the output to achieve the optimal precision, and terminating the training process when the error between the actual output and the estimated output is less than a preset value or the iteration times exceed the maximum times.
Preferably, the trained prediction model is installed on the vehicle in a sensor mode, track data transmitted by the road side laser radar is received, the road passing behavior of the pedestrian is predicted in advance, and early warning is sent out to warn the driver.
The invention has the beneficial effects that:
1. the method is based on roadside lidar point cloud data, provides prediction of the intention of pedestrians crossing/not crossing roads at different positions, has higher prediction precision compared with the traditional neural network model, and reduces the influence caused by possible failure of a vehicle-mounted sensing system when the vehicle-mounted sensing system is blocked by other vehicles and roadside objects.
2. The invention is used as a basic element for developing an infrastructure-based system, and sends warning to road users through roadside equipment, thereby reducing the hidden danger of traffic accidents.
3. Compared with the intersection signal lamp, the invention improves the working efficiency of the traffic intersection.
4. The method has high accuracy for predicting whether the pedestrian crosses the road, ensures the safety of the pedestrian, reduces the risk of traffic accidents, and improves the road passing efficiency.
Drawings
FIG. 1 is a schematic diagram of the area division and sample path of the present invention;
FIG. 2 is an architectural diagram of a depth self-encoder of the present invention;
fig. 3 is an architecture diagram of the artificial neural network of the present invention.
Detailed Description
The present invention will be further described by way of examples, but not limited thereto, with reference to the accompanying drawings.
Example 1:
as shown in fig. 1-3, the present embodiment provides a method for tracking and predicting a pedestrian crossing trajectory based on a roadside lidar, which includes the following steps:
(1) acquiring track data;
(2) preprocessing the track data acquired in the step (1);
(3) training an artificial neural network model through the track data preprocessed in the step (2);
(4) and preprocessing the trajectory data to be predicted and inputting the trajectory data into the trained artificial neural network model to obtain a prediction result.
In the step (1), historical track data is extracted from pedestrian track data of a laser radar, and data streams are converted into real-time tracks of road users through background filtering, lane recognition, road user clustering, object classification and tracking, wherein the real-time tracks comprise XYZ coordinate positions (coordinates in a data space of the laser radar), speed and direction information;
in step (2), a linear boundary is used to divide the road area (lane) and the sidewalk area, as shown in fig. 1, and a point C on the boundary represents the starting point of the crosswalk. Along the sidewalk, the area is divided into rectangular areas with the same width, and the vertical distance from the pedestrian track point to the side of the sidewalk in the rectangular area is the distance from the pedestrian to the pedestrian crossing along the sidewalk direction.
Not all crossing points occur on the crosswalk and some pedestrians choose to cross the road before reaching the crosswalk. The rectangular area is to be able to cover all pedestrians at different positions of the intersection.
The label determination standard is that the distance between the track point of the pedestrian at the last position of each area and the boundary of the sidewalk is determined, if the distance is less than 0.3m, the label is 1, the pedestrian is considered to pass through, if the track point is displayed on one side of the humanoid crosswalk of the road, the pedestrian enters the road, prediction is not needed, and the label is 1.
The distance threshold of 0.3m is determined by the average speed of the pedestrian and the rotation frequency of the lidar sensor (or the duration of one frame of data). A larger distance threshold may result in a false classification of a pedestrian not crossing the road as a pedestrian crossing the road, and a smaller distance threshold may result in a false classification of a pedestrian not crossing the road as a pedestrian crossing the road.
In fig. 2, there are four sample paths. Path 1 passes through regions R3, R2, and R1 with no traversal behavior, so the data tag of each region is equal to 0. Route 2 passes through region R3 and region R2, and finally crosses the road of region R1, so the data label for region 1 is 1, and the track points for region R2 and region R3 are labeled 0. A similar situation applies to path 3, whose label 1 represents the data in region R2. Path 4 shows an unsafe condition, with all data labeled 1 and stored in the warning dataset.
In the step (3), six characteristics of the XYZ position, the speed, the direction and the distance from the track point to the pedestrian crossing are used as input of a depth self-encoder, the six characteristics are processed by a depth self-encoder to obtain codes, then the codes and label information are input into an artificial neural network, a prediction model is trained, and the trained prediction model is used for performing label distribution on a track to be predicted.
In the step (4), the specific steps of predicting the track by the artificial neural network are as follows:
4.1: constructing an artificial neural network for labeling the track;
4.2: making a data set, wherein the data set is divided into a training set, a verification set and a test set;
4.3: the artificial neural network for labeling the trajectory is trained through a training set, and the artificial neural network for labeling the trajectory is verified through a verification set. The verification process comprises the following steps: and (3) verifying the trajectory data in the verification set by using the artificial neural network trained by the training set, and if the verification effect is not ideal, continuing training by using the training set until the requirements are met. The trained artificial neural network for labeling the track can be obtained through the steps;
4.4: and inputting the test set into a trained artificial neural network for predicting the track to obtain a classification result, wherein the classification result comprises a road passing and a road not passing.
The artificial neural network for labeling the track comprises an input layer, a hidden layer and an output layer;
the input layer is used for inputting the point cloud picture into an artificial neural network for labeling the track; the hidden layer comprises a plurality of layers of neural networks, convolution operation is carried out on the input point cloud image, point cloud information characteristics (the point cloud information characteristics of pedestrians comprise the point cloud number, the length, the width, the height and the density of a point cloud cluster) are extracted, and final overall characteristics (the shallow layer characteristics are a plurality of sections of lines formed by appearance contour points of the point cloud cluster, and the overall characteristics are the overall contours formed by combining the sections) are obtained through multiple times of convolution in the hidden layer, so that the tracks are labeled;
the output layer comprises two neurons, namely a label 1 and a label 0; the output layer is connected with the last layer of neural network of the hidden layer by adopting a full connection method, the probability of whether the pedestrian passes the road or not is obtained through parameter operation between the neurons, and the higher probability is used as a classification result.
And mounting the trained prediction model on the vehicle in a sensor mode, receiving track data transmitted by the road side laser radar, predicting the road crossing behavior of the pedestrian in advance, and giving out early warning to warn a driver.
Example 2:
a method for tracking and predicting pedestrian crossing tracks based on roadside lidar, the operation steps are as described in embodiment 1, except that in order to avoid the overfitting phenomenon of the network, a random factor is added to neurons in each layer of neural network, and in a certain layer of neural network, the random factor randomly inhibits some neurons in the layer of neural network.
The process of adding random factor inhibition is as follows: in a certain layer of neurons, a plurality of neurons are randomly extracted, and the connection between the neurons and other neurons is cut off, namely, the neurons in the previous layer skip the neurons and are directly connected with the neurons in the next layer. For example, if there are 100 neurons in a layer and the random factor is 0.05, then 100 × 0.05 — 5 neurons are selected in that layer. The selected 5 neurons are inhibited, and the upper layer and the lower layer jump over the neurons to be directly connected. Random factors are introduced between each layer of neurons.
Even if it does not function, the neurons of the upper or lower layer connected to the neuron are not affected. This is because in the hidden layer, neurons in each layer are all connected, and when a neuron is inhibited, neurons in adjacent layers are not affected.
Example 3:
a method for tracking and predicting a pedestrian crossing track based on a roadside laser radar comprises the following operation steps as in embodiment 1, and is different in that step 4.4 further comprises comparison between actual output and estimated output (actual behavior of a target to be detected), and the specific implementation process is as follows: and feeding back the error between the actual output and the estimated output to the input layer again to serve as a guide for adjusting the weight in the next round of training, adjusting the weight of each neuron in each layer of neural network through the iterative process, gradually learning the intrinsic relation between the input and the output to achieve the optimal precision, and terminating the training process when the error between the actual output and the estimated output is less than a preset value or the iteration times exceed the maximum times.
Claims (10)
1. A pedestrian crossing track tracking and predicting method based on a roadside laser radar is characterized by comprising the following operation steps:
(1) acquiring track data;
(2) preprocessing the track data acquired in the step (1);
(3) training an artificial neural network model through the track data preprocessed in the step (2);
(4) and preprocessing the trajectory data to be predicted and inputting the trajectory data into the trained artificial neural network model to obtain a prediction result.
2. The roadside lidar based pedestrian crossing trajectory tracking and predicting method according to claim 1, wherein in step (1), historical trajectory data is extracted from the pedestrian trajectory data of the lidar, and the data stream is converted into real-time trajectories of road users including XYZ coordinate position, speed, direction information by background filtering, lane recognition, road user clustering, object classification, tracking.
3. The method for tracking and predicting the street crossing trajectory of the pedestrian based on the roadside lidar as claimed in claim 1, wherein in the step (2), the linear boundary is used for dividing the road area and the sidewalk area, the area is divided into rectangular areas with the same width along the sidewalk, and the vertical distance from the pedestrian track point to the sidewalk side of the rectangular area is the distance from the pedestrian to the pedestrian crossing in the sidewalk direction.
4. The method for tracking and predicting the street crossing trajectory of the pedestrian based on the roadside lidar as claimed in claim 3, wherein the label determination criterion is that the distance between the track point of the pedestrian at the last position of each area and the boundary of the sidewalk is determined, if the distance is less than 0.3m, the label is 1, the pedestrian is considered to be crossing, if the track point is displayed on the side of the crosswalk of the road, the pedestrian is indicated to enter the road, and the label is 1 without prediction.
5. The method for tracking and predicting the pedestrian crossing trajectory based on the roadside lidar as claimed in claim 1, wherein in the step (3), the six characteristics of the XYZ position, the speed, the direction and the distance from the track point to the pedestrian crossing are used as the input of a depth self-encoder, the six characteristics are processed by the depth self-encoder to obtain a code, then the code and label information are input into an artificial neural network, a prediction model is trained, and the trained prediction model is used for performing label distribution on the trajectory to be predicted.
6. The method for tracking and predicting the pedestrian crossing trajectory based on the roadside lidar according to claim 1, wherein in the step (4), the specific steps of the artificial neural network for predicting the trajectory are as follows:
4.1: constructing an artificial neural network for labeling the track;
4.2: making a data set, wherein the data set is divided into a training set, a verification set and a test set;
4.3: training through the training set is used for carrying out the artificial neural network of label to the orbit to carry out the artificial neural network of label through verifying the collection and verifying the track, the verification process is: verifying the trajectory data in the verification set by using the artificial neural network trained by the training set, if the verification effect is not ideal, continuing training by using the training set until the requirement is met, and obtaining the trained artificial neural network for labeling the trajectory through the steps;
4.4: and inputting the test set into a trained artificial neural network for predicting the track to obtain a classification result, wherein the classification result comprises a road passing and a road not passing.
7. The roadside lidar-based pedestrian crossing trajectory tracking and prediction method of claim 6, wherein the artificial neural network for labeling the trajectory comprises an input layer, a hidden layer, an output layer;
the input layer is used for inputting the point cloud picture into an artificial neural network for labeling the track; the hidden layer comprises a plurality of layers of neural networks, convolution operation is carried out on the input point cloud picture, point cloud information characteristics are extracted, and final overall characteristics are obtained through carrying out multiple times of convolution in the hidden layer, so that the track is labeled;
the output layer comprises two neurons, namely a label 1 and a label 0; the output layer is connected with the last layer of neural network of the hidden layer by adopting a full connection method, the probability of whether the pedestrian passes the road or not is obtained through parameter operation between the neurons, and the higher probability is used as a classification result.
8. The roadside lidar-based pedestrian crossing trajectory tracking and predicting method of claim 7, wherein a stochastic factor is added to the neurons in each layer of neural network, and in a certain layer of neural network, the stochastic factor randomly suppresses certain neurons in the layer of neural network;
the process of adding random factor inhibition is as follows: in a certain layer of neurons, a plurality of neurons are randomly extracted, and the connection between the neurons and other neurons is cut off, namely, the neurons in the previous layer skip the neurons and are directly connected with the neurons in the next layer.
9. The method for tracking and predicting the pedestrian crossing trajectory based on the roadside lidar as recited in claim 8, wherein the step 4.4 further comprises comparing the actual output with the estimated output, and the specific implementation process is as follows: and feeding back the error between the actual output and the estimated output to the input layer again to serve as a guide for adjusting the weight in the next round of training, adjusting the weight of each neuron in each layer of neural network through the iterative process, gradually learning the intrinsic relation between the input and the output to achieve the optimal precision, and terminating the training process when the error between the actual output and the estimated output is less than a preset value or the iteration times exceed the maximum times.
10. The method according to claim 9, wherein the trained predictive model is installed on the vehicle in the form of a sensor, receives the trajectory data transmitted from the roadside lidar, predicts the pedestrian's road-crossing behavior in advance, and gives an early warning to warn the driver.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111312367.3A CN114067552A (en) | 2021-11-08 | 2021-11-08 | Pedestrian crossing track tracking and predicting method based on roadside laser radar |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111312367.3A CN114067552A (en) | 2021-11-08 | 2021-11-08 | Pedestrian crossing track tracking and predicting method based on roadside laser radar |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114067552A true CN114067552A (en) | 2022-02-18 |
Family
ID=80274085
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111312367.3A Withdrawn CN114067552A (en) | 2021-11-08 | 2021-11-08 | Pedestrian crossing track tracking and predicting method based on roadside laser radar |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114067552A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114997297A (en) * | 2022-05-26 | 2022-09-02 | 哈尔滨工业大学 | Target movement intention reasoning method and system based on multistage regional division |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111429754A (en) * | 2020-03-13 | 2020-07-17 | 南京航空航天大学 | Vehicle collision avoidance track risk assessment method under pedestrian crossing working condition |
| CN111444767A (en) * | 2020-02-25 | 2020-07-24 | 华中科技大学 | Pedestrian detection and tracking method based on laser radar |
| CN112487905A (en) * | 2020-11-23 | 2021-03-12 | 北京理工大学 | Method and system for predicting danger level of pedestrian around vehicle |
| CN112581791A (en) * | 2019-09-27 | 2021-03-30 | 英特尔公司 | Potential collision warning system based on road user intention prediction |
| WO2021062595A1 (en) * | 2019-09-30 | 2021-04-08 | Beijing Voyager Technology Co., Ltd. | Systems and methods for predicting a pedestrian movement trajectory |
| CN113345237A (en) * | 2021-06-29 | 2021-09-03 | 山东高速建设管理集团有限公司 | Lane-changing identification and prediction method, system, equipment and storage medium for extracting vehicle track by using roadside laser radar data |
| CN113361491A (en) * | 2021-07-19 | 2021-09-07 | 厦门大学 | Method for predicting pedestrian crossing intention of unmanned automobile |
-
2021
- 2021-11-08 CN CN202111312367.3A patent/CN114067552A/en not_active Withdrawn
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112581791A (en) * | 2019-09-27 | 2021-03-30 | 英特尔公司 | Potential collision warning system based on road user intention prediction |
| WO2021062595A1 (en) * | 2019-09-30 | 2021-04-08 | Beijing Voyager Technology Co., Ltd. | Systems and methods for predicting a pedestrian movement trajectory |
| CN111444767A (en) * | 2020-02-25 | 2020-07-24 | 华中科技大学 | Pedestrian detection and tracking method based on laser radar |
| CN111429754A (en) * | 2020-03-13 | 2020-07-17 | 南京航空航天大学 | Vehicle collision avoidance track risk assessment method under pedestrian crossing working condition |
| CN112487905A (en) * | 2020-11-23 | 2021-03-12 | 北京理工大学 | Method and system for predicting danger level of pedestrian around vehicle |
| CN113345237A (en) * | 2021-06-29 | 2021-09-03 | 山东高速建设管理集团有限公司 | Lane-changing identification and prediction method, system, equipment and storage medium for extracting vehicle track by using roadside laser radar data |
| CN113361491A (en) * | 2021-07-19 | 2021-09-07 | 厦门大学 | Method for predicting pedestrian crossing intention of unmanned automobile |
Non-Patent Citations (1)
| Title |
|---|
| JUNXUAN ZHAO等: "Trajectory tracking and prediction of pedestrian\'s crossing intention using roadside LiDAR", IET INTELLIGENT TRANSPORT SYSTEMS * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114997297A (en) * | 2022-05-26 | 2022-09-02 | 哈尔滨工业大学 | Target movement intention reasoning method and system based on multistage regional division |
| CN114997297B (en) * | 2022-05-26 | 2024-05-03 | 哈尔滨工业大学 | Target movement intention reasoning method and system based on multi-level region division |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11885910B2 (en) | Hybrid-view LIDAR-based object detection | |
| Zhao et al. | Detection and tracking of pedestrians and vehicles using roadside LiDAR sensors | |
| US10803328B1 (en) | Semantic and instance segmentation | |
| JP6591842B2 (en) | Method and system for performing adaptive ray-based scene analysis on semantic traffic space, and vehicle comprising such a system | |
| Al-qaness et al. | An improved YOLO-based road traffic monitoring system | |
| US20180349746A1 (en) | Top-View Lidar-Based Object Detection | |
| CN101966846B (en) | Travel's clear path detection method for motor vehicle involving object deteciting and enhancing | |
| US20190310651A1 (en) | Object Detection and Determination of Motion Information Using Curve-Fitting in Autonomous Vehicle Applications | |
| US20180068206A1 (en) | Object recognition and classification using multiple sensor modalities | |
| CN113345237A (en) | Lane-changing identification and prediction method, system, equipment and storage medium for extracting vehicle track by using roadside laser radar data | |
| Chen et al. | Architecture of vehicle trajectories extraction with roadside LiDAR serving connected vehicles | |
| CN107423679A (en) | A kind of pedestrian is intended to detection method and system | |
| CN116685874A (en) | Camera-laser radar fusion object detection system and method | |
| Zhang et al. | A framework for turning behavior classification at intersections using 3D LIDAR | |
| CN114093165A (en) | Roadside laser radar-based vehicle-pedestrian conflict automatic identification method | |
| Broughton et al. | Learning to see through the haze: Multi-sensor learning-fusion system for vulnerable traffic participant detection in fog | |
| CN116142233A (en) | Carrier lamp classification system | |
| CN114067552A (en) | Pedestrian crossing track tracking and predicting method based on roadside laser radar | |
| Zhao | Exploring the fundamentals of using infrastructure-based LiDAR sensors to develop connected intersections | |
| CN115713687A (en) | Method, system, and medium for determining dynamic parameters of an object | |
| Sharma et al. | Deep Learning-Based Object Detection and Classification for Autonomous Vehicles in Different Weather Scenarios of Quebec, Canada | |
| Li et al. | Composition and application of current advanced driving assistance system: A review | |
| Wu | Data processing algorithms and applications of LiDAR-enhanced connected infrastructure sensing | |
| Zyner | Naturalistic driver intention and path prediction using machine learning | |
| Moayed et al. | Surveillance-based collision-time analysis of road-crossing pedestrians |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20220218 |