CN110228413A - Oversize vehicle avoids pedestrian from being involved in the safety pre-warning system under vehicle when turning - Google Patents
Oversize vehicle avoids pedestrian from being involved in the safety pre-warning system under vehicle when turning Download PDFInfo
- Publication number
- CN110228413A CN110228413A CN201910494515.4A CN201910494515A CN110228413A CN 110228413 A CN110228413 A CN 110228413A CN 201910494515 A CN201910494515 A CN 201910494515A CN 110228413 A CN110228413 A CN 110228413A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- information
- layer
- pedestrian
- warning
- 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.)
- Granted
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60Q—ARRANGEMENT OF SIGNALLING OR LIGHTING DEVICES, THE MOUNTING OR SUPPORTING THEREOF OR CIRCUITS THEREFOR, FOR VEHICLES IN GENERAL
- B60Q9/00—Arrangement or adaptation of signal devices not provided for in one of main groups B60Q1/00 - B60Q7/00, e.g. haptic signalling
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B62—LAND VEHICLES FOR TRAVELLING OTHERWISE THAN ON RAILS
- B62D—MOTOR VEHICLES; TRAILERS
- B62D13/00—Steering specially adapted for trailers
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B62—LAND VEHICLES FOR TRAVELLING OTHERWISE THAN ON RAILS
- B62D—MOTOR VEHICLES; TRAILERS
- B62D15/00—Steering not otherwise provided for
- B62D15/02—Steering position indicators ; Steering position determination; Steering aids
- B62D15/029—Steering assistants using warnings or proposing actions to the driver without influencing the steering system
Landscapes
- Engineering & Computer Science (AREA)
- Mechanical Engineering (AREA)
- Chemical & Material Sciences (AREA)
- Combustion & Propulsion (AREA)
- Transportation (AREA)
- Human Computer Interaction (AREA)
- Traffic Control Systems (AREA)
Abstract
The present invention relates to avoid pedestrian from being involved in the safety pre-warning system under vehicle, including range radar sensor, camera, rotary angle transmitter, information integration unit, road hazard area division unit and warning device when a kind of turning of oversize vehicle.Range radar sensor acquires car body surrounding objects range information, and camera acquires vehicle two sides image, identifies with convolutional neural networks algorithm to image.Rotary angle transmitter acquires vehicle steering angle information, is divided into danger zone, prewarning area or safety zone using K mean cluster parser.Whether information integration unit merges with data information progress information of the road hazard area division unit to acquisition, handles analysis, judge with the presence of pedestrian in danger zone, and judging result is output to warning device and carries out early warning.Whether the present invention detects the environment of oversize vehicle, judge with the presence of pedestrian in danger zone, and judging result is transported to warning device and carries out early warning, improves the safety of road traffic.
Description
Technical field
The invention belongs to traffic safety technical fields, are related to a kind of pedestrains safety source of early warning, and in particular to a kind of
Oversize vehicle avoids pedestrian from being involved in the safety pre-warning system under vehicle when turning.
Background technique
Currently, oversize vehicle depositing the problems such as excessive and Driver Vision blind area poor due to turning lubrication groove in turning
So that the accident that the pedestrian of vehicle two sides is involved in vehicle bottom is frequently occurred.There are pedestrian or cycling during turning beside vehicle
Personnel, if driver or pedestrian not note that be easy the personnel of pedestrian or cycling being involved in vehicle bottom, once pedestrian is involved in vehicle
Bottom greatly may cause the injures and deaths of personnel, in some instances it may even be possible to threat to life.
The application for a patent for invention of 108791166 A of Publication No. CN discloses " a kind of to have in turning road protection pedestrian
Semitrailer, including trailer plate and rectangular block, including trailer plate and rectangular block, the left and right sides of trailer plate be equipped with accommodating groove, receive
The left and right inner wall of slot received is equipped with shifting chute, and the left and right ends of rectangular block are fixedly connected with movable plate, and movable plate and shifting chute slide
The rear end of connection, rectangular block is connect by retaining mechanism with accommodating groove rear wall, and the lower end of rectangular block is equipped with the first storage groove, the
It is connect inside one storage groove with protection mechanism ".The patent application by the two sides of semitrailer be arranged protection mechanism, by pedestrian every
From in the safety zone of semitrailer, the risk being drawn under the car is avoided.But pedestrian and driver are not given in the patent application
Dual early warning is not suitable for accident caused by the poor excessive and Driver Vision blind area of oversize vehicle turning lubrication groove.
Summary of the invention
The object of the present invention is to provide a kind of oversize vehicle turning when avoid pedestrian from being involved in the safety pre-warning system under vehicle,
The environment on left and right vehicle wheel both sides is detected when turning, identifies surrounding hazardous environment, gives driver and two sides pedestrains safety
Early warning, support vehicles safety turning and vehicle periphery pedestrains safety, improve the safety of road traffic.
The technical scheme is that oversize vehicle avoids pedestrian from being involved in the safety pre-warning system under vehicle when turning, including
Range radar sensor, camera, rotary angle transmitter, information integration unit, road hazard area division unit and warning device.
Range radar sensor acquires car body surrounding objects range information, collected target range information input to information integration unit
In be analyzed and processed.Camera acquires vehicle two sides image, identifies with convolutional neural networks algorithm to image, judges
Whether the target that range radar sensor is identified is pedestrian, and obtains the position data of the pedestrian in identification image, is passed
It is sent to information integration unit.Rotary angle transmitter acquisition vehicle steering angle information is stated, road hazard area division unit is according to corner
The collected information of sensor and automotive performance parameter are drawn the region of oversize vehicle two sides using K mean cluster parser
It is divided into danger zone, prewarning area or safety zone.Information integration unit is with road hazard area division unit to the number of acquisition
It is believed that breath carries out information fusion, processing analysis, judge whether with the presence of pedestrian in danger zone, and judging result is output to report
Alarm device.Judging result is dangerous if it exists, and driver is reminded in warning device alarm, and dangerous if it does not exist, warning device is motionless
Make.
Range radar sensor is mounted on the car body within the scope of the blind area of vehicle, and range radar sensor passes through vehicle two
Border region emits radio wave, utilizes the target echo received and the time delay measurement objects ahead of transmitting signal
Location information.Camera is mounted on above the car body of the blind area range of vehicle.Rotary angle transmitter is mounted on the side below steering wheel
Into column.Information integration unit includes data acquisition module, and data processing module, warning information analysis module and warning information are defeated
Module out.The output information of data collecting module collected range radar sensor, rotary angle transmitter and camera, information include danger
The distance between danger zone domain inner periphery target and car body, vehicle steering angle and ambient condition information.Data processing module is by ranging
The location information of the collected surrounding objects of radar sensor is matched with the collected target position information of camera, will be turned
It is imported in road area dividing system to angle information and carries out region division.Warning information analysis module is to where Vehicle target
Whether the judging result and target of danger zone are that the judging result of pedestrian is analyzed, and determine warning device under current state
Respond grade.Warning information output module is exported level results are responded to warning device.Automotive performance parameter includes that tire is disconnected
Face width, wheel outer width, wheelbase, it is preposition away from preceding foreign steamer corner.
Grade classification is carried out using region of the K mean cluster parser to vehicle two sides, is divided into danger zone, precautionary areas
Domain and safety zone three parts obtain three offline cluster mass centers by clustering offline.K mean cluster analytic process is as follows:
1) cluster number K=3 is determined as needed；
2) initialization cluster mass center, equidistantly to take a determining initial clustering mass center；
3) greatest iteration step number and centroid offset: △ d=0.0001, J=1000 are set；
Wherein: △ d is centroid offset；J is greatest iteration step number；
4) Euclidean distance between each object and cluster mass center is calculated, is divided these objects according to the numerical value of Euclidean distance
It does not incorporate into the most similar cluster therewith；
5) according to cluster result, the respective mass center of 3 clusters is recalculated, calculation method is that each respective arithmetic of cluster is flat
Number formulary；
6) 3,4 steps are repeated, until cluster result is not changing；
7) result is exported.
Euclidean distance calculation formula are as follows:
Assuming that two n-dimensional vectors are A=(a1, a2, a3……an) and B=(b1、b2、b3、……bn), then between A and B
Euclidean distance are as follows:
In formula: ρ (A, B) is Euclidean distance；
a1, a2, a3……anFor the n dimension of A；
The range of each hierarchical region of vehicle periphery road environment is determined using the above method.
Camera application convolutional neural networks (CNN) algorithm carries out the process of information processing by six layers to acquired image
It constitutes:
First layer is input layer, the image of the collected vehicle two sides of the camera of input；
The second layer includes two independent convolutional layer C1, C2, and convolution nucleus number is 40；Convolutional layer C1 convolution kernel size is
3 × 3, size is 12 × 12 × 40；Convolutional layer C2 convolution kernel size is 7 × 7, and size is 10 × 10 × 40；Convolutional layer passes through volume
Product operation extracts feature to input picture, and convolution kernel is equivalent to filter, l layers of Convolution Formula are as follows:
In formula:: l is the number of plies, xm l-1For the input of the l-1 hidden layer, n, m are two-dimensional matrix value, wn,m lFor first of hidden layer
Mapping weight matrix, bi lFor the bias matrix of first of hidden layer, yn lFor the input picture of input layer, f is activation primitive, is used
ReLU function does Nonlinear Mapping, expression formula are as follows:
Y=max (0, x)
In formula: y is output, and x is input；
One layer is pond layer behind each convolutional layer, and together by the characteristic aggregation of different location in image, realizing reduces
Characteristic dimension, and second extraction is carried out to characteristics of image, the mode in maximum pond is selected, input picture is divided in maximum pond
At nonoverlapping matrix, each subregion all exports maximum value；
Third layer includes two independent convolutional layer C3, C4, and convolution nucleus number is 65；Convolutional layer C1 convolution kernel size is
3 × 3, size is 5 × 5 × 65；Convolutional layer C2 convolution kernel size is 3 × 3, and size is that 4 × 4 × 65. 4 convolutional layer step-lengths are
2；Convolution Formula, mapping method and pond formula are identical as the second layer；
4th layer is full articulamentum F1, this layer design has 300 neurons, and the input of F1 comes from C1, C2, C3, C4；Quan Lian
Connecing layer can be enhanced network non-linear mapping capability and limitation network size size；The calculation formula of the full articulamentum of F1 are as follows:
In formula: l is the number of plies, xi l-1For the input of the l-1 hidden layer, i, j are two-dimensional matrix value, wj,i lFor first hidden layer
Map weight matrix, bj lFor the bias matrix of first of hidden layer；
Layer 5 is full articulamentum F2, and calculation formula is identical as F1；F1, F2 are using full connection；
Layer 6 is output layer, judges to classify by softmax function, softmax function calculation formula are as follows:
Xi=wix+b
In formula: θ is parameter vector, and i, j are two-dimensional matrix value, and T is temperature parameter, when T is very big, that is, tends to be just infinite
When, the corresponding activation probability of all activation values levels off to identical；When T is very low, that is, when tending to 0, different activation values is corresponding
Activate probability difference also just very big；
It completes convolutional neural networks and identifies pedestrian's process.
Warning device includes interior warning device and the outer warning device of vehicle, and interior warning device is alarmed by warning lamp and car
Device composition, it is dangerous outside vehicle for reminding driver to pay attention to.The outer warning device of vehicle is the outer alarm of vehicle, for reminding goers outside vehicle to infuse
Meaning evacuation vehicle.According to the danger zone grade where pedestrian, warning device issues vehicle different brackets alarm signal, different brackets
Alarm signal, the intensity of sound of the brightness of alarm lamp and alarm is different.
Oversize vehicle of the present invention turn when avoid pedestrian from being involved under vehicle safety pre-warning system by range radar sensor,
Camera and rotary angle transmitter detect the environment on left and right vehicle wheel both sides in turning, identify surrounding hazardous environment, input
Information fusion, processing analysis are carried out into information integration unit and road hazard area dividing system, are judged in danger zone
Whether with the presence of pedestrian, and judging result is input to warning device and carries out early warning, it is alert to driver and two sides pedestrains safety
Show, has ensured vehicle safety turning and vehicle periphery pedestrains safety, improved the safety of road traffic.The present invention has in real time
Ambient enviroment is with the presence or absence of danger when effect ground prediction oversize vehicle turning, to solve the problems, such as that oversize vehicle turning security provides one
The new thinking of kind.
Detailed description of the invention
Fig. 1 is the schematic diagram for the safety alarm device for avoiding pedestrian from being involved under vehicle when oversize vehicle of the present invention is turned；
Fig. 2 is alarm level schematic diagram；
Wherein: 1-rotary angle transmitter, 2-range radar sensors, 3-cameras, 4-road hazard region division lists
Member, 5-integrated control units, 6-warning devices, 7-data acquisition modules, 8-data processing modules, 9-warning information point
Analyse module, 10-warning information output modules, 11-interior alarms, 12-warning lamps, the outer alarm of 13-vehicles.
Specific embodiment
Below with reference to embodiment and attached drawing, the present invention is described in detail.The scope of protection of the present invention is not limited to the embodiment,
Those skilled in the art make any change within the scope of the claims and also belong to the scope of protection of the invention.
The safe early warning system that oversize vehicle of the present invention avoids pedestrian from being involved under vehicle when turning is as shown in Figure 1, include ranging thunder
Up to sensor 2, camera 3, rotary angle transmitter 1, information integration unit 5, road hazard area division unit 4 and warning device 6.
Information integration unit 5 includes data acquisition module 7, data processing module 8, warning information analysis module 9 and warning information output
Module 10.Warning device 6 includes interior warning device and the outer warning device of vehicle, and interior warning device is reported by warning lamp 12 and car
Alert device 11 forms, dangerous outside vehicle for reminding driver to pay attention to.The outer warning device of vehicle is the outer alarm 13 of vehicle, for reminding outside vehicle
Pedestrian pays attention to avoiding vehicle.Range radar sensor and camera and data acquisition module communication connection, rotary angle transmitter pass through
Road hazard area division unit and data acquisition module communication connection.Road hazard area division unit is joined equipped with automotive performance
Number input terminals, automotive performance parameter include deflected width of tyre, wheel outer width, wheelbase, it is preposition away from preceding foreign steamer corner.Data are adopted
Collect module successively with data processing module, warning information analysis module and warning information output module communication connection, warning information
Output module and warning device communication connection.Range radar sensor acquires car body surrounding objects range information, collected mesh
Subject distance information input is analyzed and processed into information integration unit.Camera acquires vehicle two sides image, with convolution mind
Identify judge whether the target that range radar sensor is identified is pedestrian, and identified to image through network algorithm
The position data of pedestrian in image is transmitted to information integration unit.Rotary angle transmitter acquires vehicle steering angle information, road
Road risk zontation unit is analyzed according to the collected information of rotary angle transmitter and automotive performance parameter using K mean cluster
The region division of oversize vehicle two sides is danger zone, prewarning area or safety zone by algorithm.Information integration unit and road
Risk zontation unit carries out information fusion to the data information of acquisition, processing is analyzed, and judges whether there is row in danger zone
People exists, and judging result is output to warning device.Judging result is dangerous if it exists, and driver is reminded in warning device alarm,
Dangerous if it does not exist, warning device is failure to actuate.
Range radar sensor 2 is mounted on the car body within the scope of the blind area of vehicle, and range radar sensor passes through vehicle
Areas at both sides emits radio wave, utilizes the time delay measurement objects ahead of the target echo and transmitting signal that receive
Location information.Camera 3 is mounted on above the car body of the blind area range of vehicle, and rotary angle transmitter 1 is mounted below steering wheel
Steering column in.
Data acquisition module 7 acquires the output information of range radar sensor, rotary angle transmitter and camera, and information includes
The distance between danger zone inner periphery target and car body, vehicle steering angle and ambient condition information.Data processing module 8 will be surveyed
Surrounding objects location information away from radar sensor acquisition is matched with the collected target position information of camera, will be turned to
Angle information imports in road area dividing system and carries out region division.Warning information analysis module 9 is to where Vehicle target
Whether the judging result and target of danger zone are that the judging result of pedestrian is analyzed, and determine warning device under current state
Respond grade.Warning information output module 10 is exported level results are responded to warning device 6.
The purpose clustered offline is to carry out grade classification to vehicle areas at both sides, obtains offline cluster matter by clustering offline
The heart.The collected information of rotary angle transmitter and automotive performance parameter are input to the road danger using K mean cluster parser
Dangerous area division unit 4, by oversize vehicle two sides by the region division of vehicle two sides be danger zone, prewarning area and safety
Region three parts obtain three offline cluster mass centers by clustering offline.K mean cluster analytic process is as follows:
1) cluster number K=3 is determined as needed；
2) initialization cluster mass center, equidistantly to take a determining initial clustering mass center；
3) greatest iteration step number and centroid offset: △ d=0.0001, J=1000 are set；
Wherein: △ d is centroid offset；J is greatest iteration step number；
4) Euclidean distance between each object and cluster mass center is calculated, is divided these objects according to the numerical value of Euclidean distance
It does not incorporate into the most similar cluster therewith；
5) according to cluster result, the respective mass center of 3 clusters is recalculated, calculation method is that each respective arithmetic of cluster is flat
Number formulary；
6) 3,4 steps are repeated, until cluster result is not changing；
7) result is exported.
Euclidean distance calculation formula are as follows:
Assuming that two n-dimensional vectors are A=(a1, a2, a3……an) and B=(b1、b2、b3、……bn), then between A and B
Euclidean distance are as follows:
ρ (A, B) is Euclidean distance in formula；
a1, a2, a3……anFor the n dimension of A；
The range of each hierarchical region of vehicle periphery road environment is determined using the above method.
Camera application convolutional neural networks (CNN) algorithm carries out the process of information processing by six layers to acquired image
It constitutes:
First layer is input layer, the image of the collected vehicle two sides of the camera of input；
The second layer includes two independent convolutional layer C1, C2, and convolution nucleus number is 40；Convolutional layer C1 convolution kernel size is
3 × 3, size is 12 × 12 × 40；Convolutional layer C2 convolution kernel size is 7 × 7, and size is 10 × 10 × 40；Convolutional layer passes through volume
Product operation extracts feature to input picture, and convolution kernel is equivalent to filter, l layers of Convolution Formula are as follows:
In formula: l is the number of plies, xm l-1For the input of the l-1 hidden layer, n, m are two-dimensional matrix value, wn,m lFor first hidden layer
Map weight matrix, bi lFor the bias matrix of first of hidden layer, yn lFor the input picture of input layer, f is activation primitive, is used
ReLU function does Nonlinear Mapping, expression formula are as follows:
Y=max (0, x)
In formula: y is output, and x is input；
One layer is pond layer behind each convolutional layer, and together by the characteristic aggregation of different location in image, realizing reduces
Characteristic dimension, and second extraction is carried out to characteristics of image, the mode in maximum pond is selected, input picture is divided in maximum pond
At nonoverlapping matrix, each subregion all exports maximum value；
Third layer includes two independent convolutional layer C3, C4, and convolution nucleus number is 65；Convolutional layer C1 convolution kernel size is
3 × 3, size is 5 × 5 × 65；Convolutional layer C2 convolution kernel size is 3 × 3, and size is that 4 × 4 × 65. 4 convolutional layer step-lengths are
2；Convolution Formula, mapping method and pond formula are identical as the second layer；
4th layer is full articulamentum F1, this layer design has 300 neurons, and the input of F1 comes from C1, C2, C3, C4；Quan Lian
Connecing layer can be enhanced network non-linear mapping capability and limitation network size size；The calculation formula of the full articulamentum of F1 are as follows:
In formula: l is the number of plies, xi l-1For the input of the l-1 hidden layer, i, j are two-dimensional matrix value, wj,i lFor first hidden layer
Map weight matrix, bj lFor the bias matrix of first of hidden layer；
Layer 5 is full articulamentum F2, and calculation formula is identical as F1；F1, F2 are using full connection；
Layer 6 is output layer, judges to classify by softmax function, softmax function calculation formula are as follows:
Xi=wix+b
In formula: θ is parameter vector, and i, j are two-dimensional matrix value, and T is temperature parameter, when T is very big, that is, tends to be just infinite
When, the corresponding activation probability of all activation values levels off to identical；When T is very low, that is, when tending to 0, different activation values is corresponding
Activate probability difference also just very big；
It completes convolutional neural networks and identifies pedestrian's process.
The course of work for the safety pre-warning system that oversize vehicle of the present invention avoids pedestrian from being involved under vehicle when turning, step is such as
Under:
(1) rotary angle transmitter 1 acquires vehicle steering angle information first, by vehicle steering angle information input to road hazard area
Domain division unit 4 divides vehicle two side areas intersexuality, is divided into danger zone, prewarning area and safety zone three parts；
(2) range radar sensor 2 utilizes the target echo received by field emission radio wave forwards
With transmitting signal time delay measurement objects ahead location information, by target range information input to information integration unit 5
In, determine current goal in which region；
(3) the image information around the acquisition oversize vehicle of camera 3 on the car body is installed, is calculated with convolutional neural networks
Image information is carried out structuring processing by method, identifies whether the target of range radar is pedestrian；
(4) the data information that information integration unit 5 is acquired with 4 Duis of road hazard area division unit carries out information and merges, locates
Whether reason analysis, judge in danger zone with the presence of pedestrian；
(5) it will determine that result is input to warning information processing module 10 and analyzes, judge whether target is pedestrian；If sentencing
Determine unidentified to pedestrian in result, then inputs alarm level information to warning device according to target region；If it is determined that result
There are pedestrians, then according to the upper level alarm level information alert of pedestrian region；
(6) warning message is transported to warning device 6, dangerous if it exists, warning device alarm reminds driver, if not depositing
In danger, warning device is failure to actuate.
Warning device of the invention uses classifying alarm, as shown in Fig. 2, when range radar and camera are judged currently
After target and target region, the response of classifying alarm device.Prior-warning device has four grades altogether, respectively level-one alarm,
Secondary alarm, three-level are alarmed and without response.If target is pedestrian, when pedestrian is in danger zone, level-one is taken to alarm；Row
People takes secondary alarm in prewarning area；Pedestrian is in safety zone, and warning device is without response.If target is not pedestrian,
Successively than pedestrian low level-one, target take secondary alarm in danger zone to alarm level；Target is taken in prewarning area
Three-level alarm, target is in safety zone, and warning device is without response.
Claims (8)
1. a kind of oversize vehicle avoids pedestrian from being involved in the safety pre-warning system under vehicle when turning, it is characterized in that: described device includes
Range radar sensor (2), camera (3), rotary angle transmitter (1), information integration unit (5), road hazard region division list
First (4) and warning device (6)；The range radar sensor acquires car body surrounding objects range information, collected target away from
It is analyzed and processed from information input into information integration unit；The camera acquires vehicle two sides image, with convolution mind
Identify judge whether the target that range radar sensor is identified is pedestrian, and identified to image through network algorithm
The position data of pedestrian in image is transmitted to information integration unit；The rotary angle transmitter acquisition vehicle steering angle letter
Breath, the road hazard area division unit is according to the collected information of rotary angle transmitter and automotive performance parameter, using K mean value
The region division of oversize vehicle two sides is danger zone, prewarning area or safety zone by cluster algorithm；The information collection
It merged at unit with data information progress information of the road hazard area division unit to acquisition, handle analysis, judge danger area
Whether with the presence of pedestrian in domain, and judging result is output to warning device；Judging result is dangerous if it exists, warning device report
It is alert, driver is reminded, dangerous if it does not exist, warning device is failure to actuate.
2. oversize vehicle according to claim 1 avoids pedestrian from being involved in the safety pre-warning system under vehicle, feature when turning
Be: on the car body that the range radar sensor (2) is mounted within the scope of the blind area of vehicle, range radar sensor passes through vehicle
Areas at both sides emits radio wave, utilizes the time delay measurement objects ahead of the target echo and transmitting signal that receive
Location information；The camera (3) is mounted on above the car body of the blind area range of vehicle；Rotary angle transmitter (1) installation
In the steering column below steering wheel.
3. oversize vehicle according to claim 1 avoids pedestrian from being involved in the safety pre-warning system under vehicle, feature when turning
Be: the information integration unit (5) includes data acquisition module (7), data processing module (8), warning information analysis module (9)
With warning information output module (10)；The data collecting module collected range radar sensor, rotary angle transmitter and camera
Output information, information include the distance between danger zone inner periphery target and car body, vehicle steering angle and ambient enviroment letter
Breath；The surrounding objects location information and the collected target of camera that the data processing module acquires range radar sensor
Location information is matched, and will be turned in angle information importing road area dividing system and be carried out region division；The warning information
Whether analysis module is that the judging result of pedestrian carries out to the judging result and target of danger zone where Vehicle target
Analysis determines that warning device responds grade under current state；The warning information output module will respond level results export to
Warning device.
4. oversize vehicle according to claim 1 avoids pedestrian from being involved in the safety pre-warning system under vehicle, feature when turning
Be: the automotive performance parameter include deflected width of tyre, wheel outer width, wheelbase, it is preposition away from preceding foreign steamer corner.
5. oversize vehicle according to claim 1 avoids pedestrian from being involved in the safety pre-warning system under vehicle, feature when turning
It is: it is described that grade classification is carried out using region of the K mean cluster parser to vehicle two sides, it is divided into danger zone, precautionary areas
Domain and safety zone three parts obtain three offline cluster mass centers by clustering offline；The K mean cluster analytic process is such as
Under:
1) cluster number K=3 is determined as needed；
2) initialization cluster mass center, equidistantly to take a determining initial clustering mass center；
3) greatest iteration step number and centroid offset: △ d=0.0001, J=1000 are set；
Wherein: △ d is centroid offset；J is greatest iteration step number；
4) Euclidean distance between each object and cluster mass center is calculated, is drawn these objects respectively according to the numerical value of Euclidean distance
It is grouped into the most similar cluster therewith；
5) according to cluster result, the respective mass center of 3 clusters is recalculated, calculation method is each respective arithmetic square of cluster
Number；
6) 3,4 steps are repeated, until cluster result is not changing；
7) result is exported.
6. oversize vehicle according to claim 5 avoids pedestrian from being involved in the safety pre-warning system under vehicle, feature when turning
It is: the Euclidean distance calculation formula are as follows:
Assuming that two n-dimensional vectors are A=(a1, a2, a3……an) and B=(b1、b2、b3、……bn), then it is European between A and B
Distance are as follows:
In formula: ρ (A, B) is Euclidean distance；
a1, a2, a3……anFor the n dimension of A；
The range of each hierarchical region of vehicle periphery road environment is determined using the above method.
7. oversize vehicle according to claim 1 avoids pedestrian from being involved in the safety pre-warning system under vehicle, feature when turning
Be: camera application convolutional neural networks (CNN) algorithm carries out the process of information processing by six layers to acquired image
It constitutes:
First layer is input layer, the image of the collected vehicle two sides of the camera of input；
The second layer includes two independent convolutional layer C1, C2, and convolution nucleus number is 40；Convolutional layer C1 convolution kernel size be 3 ×
3, size is 12 × 12 × 40；Convolutional layer C2 convolution kernel size is 7 × 7, and size is 10 × 10 × 40；Convolutional layer is grasped by convolution
Work extracts feature to input picture, and convolution kernel is equivalent to filter, l layers of Convolution Formula are as follows:
In formula: l is the number of plies, xm l-1For the input of the l-1 hidden layer, n, m are two-dimensional matrix value, wn,m lFor the mapping of first of hidden layer
Weight matrix, bi lFor the bias matrix of first of hidden layer, yn lFor the input picture of input layer, f is activation primitive, using ReLU letter
Number does Nonlinear Mapping；Expression formula are as follows:
Y=max (0, x)
In formula: y is output, and x is input；
One layer is pond layer behind each convolutional layer, and together by the characteristic aggregation of different location in image, realizing reduces feature
Dimension, and second extraction is carried out to characteristics of image, the mode in maximum pond is selected, maximum pond is divided into input picture not
The matrix of overlapping, each subregion all export maximum value；
Third layer includes two independent convolutional layer C3, C4, and convolution nucleus number is 65；Convolutional layer C1 convolution kernel size be 3 ×
3, size is 5 × 5 × 65；Convolutional layer C2 convolution kernel size is 3 × 3, and size is that 4 × 4 × 65. 4 convolutional layer step-lengths are 2；Volume
Product formula, mapping method and pond formula are identical as the second layer；
4th layer is full articulamentum F1, this layer design has 300 neurons, and the input of F1 comes from C1, C2, C3, C4；Full articulamentum
Network non-linear mapping capability and limitation network size size can be enhanced；The calculation formula of the full articulamentum of F1 are as follows:
In formula: l is the number of plies, xi l-1For the input of the l-1 hidden layer, i, j are two-dimensional matrix value, wj,i lFor the mapping of first of hidden layer
Weight matrix, bj lFor the bias matrix of first of hidden layer；
Layer 5 is full articulamentum F2, and calculation formula is identical as F1；F1, F2 are using full connection；
Layer 6 is output layer, judges to classify by softmax function, softmax function calculation formula are as follows:
Xi=wix+b
In formula: θ is parameter vector, and i, j are two-dimensional matrix value, and T is temperature parameter, when T is very big, that is, when tending to be just infinite, and institute
The corresponding activation probability of some activation values levels off to identical；When T is very low, that is, when tending to 0, the corresponding activation of different activation values is general
Rate difference is also just very big；
It completes convolutional neural networks and identifies pedestrian's process.
8. oversize vehicle according to claim 1 avoids pedestrian from being involved in the safety pre-warning system under vehicle, feature when turning
Be: the warning device (6) includes interior warning device and the outer warning device of vehicle, and the car warning device is by warning lamp (12)
It is formed with interior alarm (11), it is dangerous outside vehicle for reminding driver to pay attention to；The outer warning device of the vehicle is the outer alarm of vehicle
(13), for reminding goers outside vehicle to pay attention to avoiding vehicle；According to the danger zone grade where pedestrian, warning device issues vehicle not
Ad eundem alarm signal, different grades of alarm signal, the brightness of alarm lamp and the intensity of sound of alarm are different.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910494515.4A CN110228413B (en) | 2019-06-10 | 2019-06-10 | Safety early warning system for avoiding pedestrians from being involved under large-scale vehicle during turning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910494515.4A CN110228413B (en) | 2019-06-10 | 2019-06-10 | Safety early warning system for avoiding pedestrians from being involved under large-scale vehicle during turning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110228413A true CN110228413A (en) | 2019-09-13 |
| CN110228413B CN110228413B (en) | 2020-07-14 |
Family
ID=67859441
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910494515.4A Expired - Fee Related CN110228413B (en) | 2019-06-10 | 2019-06-10 | Safety early warning system for avoiding pedestrians from being involved under large-scale vehicle during turning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110228413B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110936937A (en) * | 2019-03-21 | 2020-03-31 | 湖南三一港口设备有限公司 | Automatic braking method and system |
| CN111047822A (en) * | 2019-12-18 | 2020-04-21 | 万翼科技有限公司 | Danger early warning equipment, danger early warning method and related products thereof |
| CN111505577A (en) * | 2020-04-27 | 2020-08-07 | 湖南大学 | Mobile vehicle positioning method based on visible light communication |
| CN111583336A (en) * | 2020-04-22 | 2020-08-25 | 深圳市优必选科技股份有限公司 | Robot and inspection method and device thereof |
| CN113212427A (en) * | 2020-02-03 | 2021-08-06 | 通用汽车环球科技运作有限责任公司 | Intelligent vehicle with advanced vehicle camera system for underbody hazard and foreign object detection |
| CN114516299A (en) * | 2022-01-21 | 2022-05-20 | 国网甘肃省电力公司综合服务中心 | Vehicle alarm system and alarm method thereof |
| CN117854046A (en) * | 2024-03-07 | 2024-04-09 | 北京理工大学前沿技术研究院 | Integrated positioning system and device based on vision fusion |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20010027363A1 (en) * | 1999-12-24 | 2001-10-04 | Kazunori Shimazaki | Steering assist apparatus |
| US20090212930A1 (en) * | 2005-03-03 | 2009-08-27 | Continental Teves Ag & Co. Ohg | Method and Device for Avoiding a Collision in a Lane Change Maneuver of a Vehicle |
| US20160291149A1 (en) * | 2015-04-06 | 2016-10-06 | GM Global Technology Operations LLC | Fusion method for cross traffic application using radars and camera |
| CN106740829A (en) * | 2017-03-23 | 2017-05-31 | 吉林大学 | Based on the double semi-dragging truck riding stability automatic identifications of cluster analysis and early warning system |
| CN107633220A (en) * | 2017-09-13 | 2018-01-26 | 吉林大学 | A kind of vehicle front target identification method based on convolutional neural networks |
| CN107862287A (en) * | 2017-11-08 | 2018-03-30 | 吉林大学 | A kind of front zonule object identification and vehicle early warning method |
| CN107985200A (en) * | 2017-11-13 | 2018-05-04 | 西南交通大学 | A kind of load truck right-hand bend safety pre-warning system and method |
| CN108629968A (en) * | 2018-06-26 | 2018-10-09 | 吉林大学 | A kind of pedestrian's street crossing safety control based on laser radar |
| CN108877261A (en) * | 2017-05-10 | 2018-11-23 | 杜卡迪摩托控股有限公司 | With the motorcycle for detecting the device from the close vehicle in rear |
| CN208498373U (en) * | 2018-05-25 | 2019-02-15 | 武汉科技大学 | The lateral anti-collision system of vehicle |
-
2019
- 2019-06-10 CN CN201910494515.4A patent/CN110228413B/en not_active Expired - Fee Related
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20010027363A1 (en) * | 1999-12-24 | 2001-10-04 | Kazunori Shimazaki | Steering assist apparatus |
| US20090212930A1 (en) * | 2005-03-03 | 2009-08-27 | Continental Teves Ag & Co. Ohg | Method and Device for Avoiding a Collision in a Lane Change Maneuver of a Vehicle |
| US20160291149A1 (en) * | 2015-04-06 | 2016-10-06 | GM Global Technology Operations LLC | Fusion method for cross traffic application using radars and camera |
| CN106740829A (en) * | 2017-03-23 | 2017-05-31 | 吉林大学 | Based on the double semi-dragging truck riding stability automatic identifications of cluster analysis and early warning system |
| CN108877261A (en) * | 2017-05-10 | 2018-11-23 | 杜卡迪摩托控股有限公司 | With the motorcycle for detecting the device from the close vehicle in rear |
| CN107633220A (en) * | 2017-09-13 | 2018-01-26 | 吉林大学 | A kind of vehicle front target identification method based on convolutional neural networks |
| CN107862287A (en) * | 2017-11-08 | 2018-03-30 | 吉林大学 | A kind of front zonule object identification and vehicle early warning method |
| CN107985200A (en) * | 2017-11-13 | 2018-05-04 | 西南交通大学 | A kind of load truck right-hand bend safety pre-warning system and method |
| CN208498373U (en) * | 2018-05-25 | 2019-02-15 | 武汉科技大学 | The lateral anti-collision system of vehicle |
| CN108629968A (en) * | 2018-06-26 | 2018-10-09 | 吉林大学 | A kind of pedestrian's street crossing safety control based on laser radar |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110936937A (en) * | 2019-03-21 | 2020-03-31 | 湖南三一港口设备有限公司 | Automatic braking method and system |
| CN110936937B (en) * | 2019-03-21 | 2021-03-30 | 湖南三一港口设备有限公司 | Automatic braking method and system |
| CN111047822A (en) * | 2019-12-18 | 2020-04-21 | 万翼科技有限公司 | Danger early warning equipment, danger early warning method and related products thereof |
| CN113212427A (en) * | 2020-02-03 | 2021-08-06 | 通用汽车环球科技运作有限责任公司 | Intelligent vehicle with advanced vehicle camera system for underbody hazard and foreign object detection |
| CN111583336A (en) * | 2020-04-22 | 2020-08-25 | 深圳市优必选科技股份有限公司 | Robot and inspection method and device thereof |
| CN111583336B (en) * | 2020-04-22 | 2023-12-01 | 深圳市优必选科技股份有限公司 | Robot and inspection method and device thereof |
| CN111505577A (en) * | 2020-04-27 | 2020-08-07 | 湖南大学 | Mobile vehicle positioning method based on visible light communication |
| CN114516299A (en) * | 2022-01-21 | 2022-05-20 | 国网甘肃省电力公司综合服务中心 | Vehicle alarm system and alarm method thereof |
| CN117854046A (en) * | 2024-03-07 | 2024-04-09 | 北京理工大学前沿技术研究院 | Integrated positioning system and device based on vision fusion |
| CN117854046B (en) * | 2024-03-07 | 2024-05-14 | 北京理工大学前沿技术研究院 | Integrated positioning system and device based on vision fusion |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110228413B (en) | 2020-07-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110228413A (en) | Oversize vehicle avoids pedestrian from being involved in the safety pre-warning system under vehicle when turning | |
| US10677894B2 (en) | Vehicle sensing system for classification of vehicle model | |
| CN108099786A (en) | A kind of driving side blind area excludes early warning system | |
| CN207697623U (en) | A kind of driving side blind area exclusion early warning system | |
| CN109153359A (en) | A kind of monitoring method and system of vehicle blind zone | |
| Perumal et al. | An insight into crash avoidance and overtaking advice systems for autonomous vehicles: A review, challenges and solutions | |
| CN105512623A (en) | Foggy-day driving visual enhancement and visibility early warning system and method based on multiple sensors | |
| CN103842844A (en) | Detection device for vehicle, abnormality detection method, and abnormality detection program | |
| CN101934771A (en) | Vehicle collision warning system | |
| CN106828304A (en) | Car backing warning method using structure light sensing obstacle | |
| CN113276770B (en) | Commercial vehicle total blind area monitoring system and method based on safety and low cost requirements | |
| CN113096387B (en) | Vehicle-mounted road surface monitoring and early warning method, system, terminal and storage medium | |
| CN110378202A (en) | One kind being based on fish-eye comprehensive pedestrian impact method for early warning | |
| CN104104915A (en) | Multifunctional driving monitoring early warning system based on mobile terminal | |
| US20210405185A1 (en) | System and method providing truck-mounted sensors to detect trailer following vehicles and trailer conditions | |
| CN106740458A (en) | A kind of enabling safety prompting system and method for vehicle | |
| EP4242062A1 (en) | Pedestrian alert system | |
| CN112793507A (en) | Blind area early warning braking system based on inertial device perception vehicle right turn motion characteristic | |
| Sotelo et al. | Blind spot detection using vision for automotive applications | |
| Bagi et al. | Sensing structure for blind spot detection system in vehicles | |
| CN106778907A (en) | A kind of intelligent travelling crane early warning system based on multi-sensor information fusion | |
| CN103177248A (en) | Rapid pedestrian detection method based on vision | |
| CN115862383B (en) | Uncertainty-based vehicle blind area potential risk quantification method | |
| CN108256418A (en) | A kind of pedestrian's method for early warning and system based on infrared imaging | |
| Meinecke et al. | Approach for protection of vulnerable road users using sensor fusion techniques |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200714 Termination date: 20210610 |