CN110110632A - A kind of the vision ADAS target detection identifying system and device of low cost - Google Patents
A kind of the vision ADAS target detection identifying system and device of low cost Download PDFInfo
- Publication number
- CN110110632A CN110110632A CN201910346928.8A CN201910346928A CN110110632A CN 110110632 A CN110110632 A CN 110110632A CN 201910346928 A CN201910346928 A CN 201910346928A CN 110110632 A CN110110632 A CN 110110632A
- Authority
- CN
- China
- Prior art keywords
- network
- driver
- vehicle
- face
- processor
- 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.)
- Pending
Links
Classifications
-
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/59—Context or environment of the image inside of a vehicle, e.g. relating to seat occupancy, driver state or inner lighting conditions
- G06V20/597—Recognising the driver's state or behaviour, e.g. attention or drowsiness
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B21/00—Alarms responsive to a single specified undesired or abnormal condition and not otherwise provided for
- G08B21/02—Alarms for ensuring the safety of persons
- G08B21/06—Alarms for ensuring the safety of persons indicating a condition of sleep, e.g. anti-dozing alarms
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N7/00—Television systems
- H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
- H04N7/181—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast for receiving images from a plurality of remote sources
Abstract
The invention discloses a kind of inexpensive vision ADAS aims of systems detection recognition method and devices, it is related to intelligent and safe and drives ancillary technique field.The visual perception method uses depth learning technology, clarification of objective extraction is carried out by designed dedicated one-stage network, face can be effectively extracted, the high layer information of the targets such as vehicle, pedestrian, newly-designed network model being capable of the accurate operation in real time on calculating the low mobile unit of power.The invention also discloses a kind of vehicle-mounted vision ADAS equipment, which uses advanced depth learning technology, can realize that real-time, accurate, stable object detection and recognition, device hardware are low in cost in built-in terminal.
Description
Technical field
The present invention relates to intelligent and safes to drive ancillary technique field, and in particular to a kind of vision ADAS target inspection of low cost
Survey identifying system and device.
Background technique
The sensing mode of ADAS (Advanced Driver Assistance Systems) rely primarily on (millimeter wave or swash
Light) radar, infrared or camera, in contrast, camera has lower cost, high accuracy, perceives week closest to human eye
The advantage of information mode is enclosed, inexpensive vision ADAS is received more and more attention.With the emergence of artificial intelligence technology, lead to
It crosses the depth learning technology ADAS that energizes and is increasingly becoming the new hot spot of industry and academia.But it is multiple to be limited to deep learning network program
It is miscellaneous, occupy that memory is big, recognition speed is slow, accuracy of identification is low, calculate the factors such as the time is longer, the reality of practical application be often not achieved
When property requirement, simultaneously as illumination, the influence of weather, different postures, the robustness of existing ADAS product are poor.
In recent years, the net about commerial vehicles such as vehicle, shared automobile, lorry constantly produce vicious behaviour safety accident, such as 2018
It allows drop to drip the two pieces being on the teeth of the storm car owner's murder with the wind, allows the safety management to net about vehicle to be promoted new to one
Height, relevant departments have put into effect the laws and regulations that special focus efforts on special areas people vehicle is not inconsistent confusion.By deep learning network application
To the low embedded end of power is calculated, relatively good effect is not yet obtained in terms of meeting real-time, accuracy, stability at present.
Summary of the invention
The object of the present invention is to provide the vision ADAS target detection identifying systems and device of a kind of low cost, by depth
Habit technology introduces ADAS, and main ADAS function is realized with machine vision.The present invention can be realized in car-mounted terminal and be driven in real time
Identification authentication, the perception of driving ambient enviroment and the ADAS early warning for the person of sailing.
The technical problem to be solved by the present invention is to how low cost vehicle-mounted embedded type terminal on realize real time execution
Deep learning frame, and there is stronger robustness to illumination and different postures.The invention is realized by the following technical scheme:
The vision ADAS target detection identifying system of low cost, it is characterised in that apply one-stage deep learning network group, institute
Stating deep learning network group mainly includes first network, the second network and third network, and the first network is interesting target
Network is detected, second network is image rectification network, and the third network is characterized coding network, is realized to interested
Clarification of objective is extracted, and feature vector is obtained.
The described method includes:
In the first network, first with data set disclosed on network, primary mold A1 is trained, further according to product reality
Border application scenarios record different illumination, the video of the interior driver under weather, background, vehicle external environment video, and to video counts
According to manually being marked, while data enhancing is carried out, be then sent through the primary mold A1 and carry out retraining and obtain model A2, instructed
The model A2 perfected can detecte out the interesting target in a picture.
In second network, first with data set disclosed on network, primary mold B1 is trained, is recycled certainly
Dynamic marking software carries out automatic marking to target, and the target image for then crossing automatic marking is sent into the primary mold B1 and carries out
Retraining obtains Model B 2.Trained Model B 2 fast and accurate can find target signature, subsequently through the rotation to picture
Turn, scaling and mistake are cut, and realize the correction to picture.Because the training of the second network carries out data mark using automatic marking software
Artificial mark cost can be greatly reduced in note.
The third network is a depth convolutional neural networks, it can be allowed to generate measured value for target by training, real
The extraction of existing target feature vector.For example in driver identification identification authentication, the trained third network can be to driving
The scene photograph of member extracts a feature vector F1, extracts a feature vector F2 to the accredited certificate photo of driver.
Then, Euclidean distance is calculated to described eigenvector F1 and F2, obtains similarity, to determine driver and net that scene is taken pictures
Whether the driver that network is registered realizes identification to driver authenticate as same people.
A kind of vision ADAS target detection identification device of low cost, which is characterized in that described device include: forward sight, after
Depending on camera, processor.The forward sight, rearview camera and processor connect；The forward sight camera is for obtaining outside vehicle
The ambient video data in portion, and the ambient video data are sent to the processor；The rearview camera is for obtaining
Driver's video data, and driver's video data is sent to the processor；The processor is used to regard the environment
Frequency will test result and be sent to warning module according to being detected and identified.
In a kind of possible design, the warning module is also used to carry out fatigue driving early warning,
In a kind of possible design, the main equipment further include: SIM card module, 4G module；
The processor is connect with the 4G module respectively, and the 4G module is connected with SIM card module；The SIM card module is used
In connection carrier network signal, the 4G module is for ADAS equipment to be connected with server.
Detailed description of the invention
Below by clearly understandable mode, preferred embodiment is described with reference to the drawings, to a kind of inexpensive vision ADAS
Above-mentioned characteristic, technical characteristic, advantage and its implementation of aims of systems detection recognition method and device are further described.
Fig. 1 is a structure diagram of the invention.
Fig. 2 is the flow chart of one embodiment of the vision ADAS target detection identifying system of low cost of the invention；
Drawing reference numeral explanation:
100. sample collection module, 200. data predictions enhance module, 300. network training modules, the identification of 400. target detections
Module.
Specific embodiment
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, Detailed description of the invention will be compareed below
A specific embodiment of the invention.It should be evident that drawings in the following description are only some embodiments of the invention, for
For those of ordinary skill in the art, without creative efforts, it can also be obtained according to these attached drawings other
Attached drawing, and obtain other embodiments.
To make simplified form, part related to the present invention is only schematically shown in each figure, they are not represented
Its practical structures as product.In addition, there is identical structure or function in some figures so that simplified form is easy to understand
Component only symbolically depicts one of those, or has only marked one of those.Herein, "one" is not only indicated
" only this ", can also indicate the situation of " more than one ".
In one embodiment of the invention, as shown in Figure 1, a kind of low cost vision ADAS aims of systems detects identification side
Method, comprising:
Step S100 first obtains sample data from the public data collection on network, then manually acquires exclusive data and mark as instruction
Practice sample data.
Specifically, first concentrating the sample for obtaining a part to carry out network pre-training from the public data on network.Then, exist
Different illumination acquire environment and human face data as much as possible under different weather, special operation background, and to these acquisitions
Data carry out artificial and automatic marking.Data enhancing is carried out to sample data simultaneously, positive negative sample balances, and difficult screening sample etc. is pre-
Processing operation, using pretreated sample as new training sample.
Step S200 design object detection network simultaneously carries out network training.
Specifically, one special one-stage network of design, first carries out pre-training to this network with public data.
Again with voluntarily acquiring and the data marked are trained, which can be realized to the interested target area of a width picture
Detection and identification.
Step S300 designs picture corrective network, is corrected to the Target Photo that S200 is detected.
Specifically, first carrying out pre-training to this network with public data.It is examined again using automatic marking software to by S200
The Target Photo measured carries out automatic marking, and the Target Photo of the small size after automatic marking is then sent into target correction network
It is trained, which can realize the correction to Target Photo, in case subsequent network is easier to detect target feature point.
Step S400 design feature coding network carries out feature extraction and coding to the Target Photo after S300 correction.
Specifically, first carrying out pre-training to this network with public data, then target is acquired in different scenes, do not share the same light
According to, the picture under different weather, different background, these samples are sent into feature coding network and are trained, the loss of the network
The purpose of design of function is to make inter- object distance minimum, and between class distance is maximum, and the prediction output of network is a feature representation vector.
In another embodiment of the present invention, as shown in Fig. 2, a kind of low cost vision ADAS aims of systems detection identification
Device, for realizing target detection recognizer as described above, comprising: main equipment.
Specifically, above-mentioned main equipment includes: processor 201, forward sight thermal infrared camera 202, backsight infrared camera
203 and alarm 204.
Specifically, rearview camera 203, for acquiring the image of vehicle body ambient enviroment；Forward sight camera 202 is used for collecting vehicle
The image of interior personnel；Processor 201 carries out detection identification to interesting target, if in advance for the operation of deep learning model
Measuring vehicle may collide in safety time with Ta Che or pedestrian, then sends alarm command to warning module；Early warning
Module 204 sends prewarning speech prompting according to the alarm command received；Cloud manages platform, management and place for data
Manage device.
Specifically, vehicle-mounted ADAS device is generally applicable in and is mounted in various types of vehicles, and can using device itself or
Be that camera on vehicle collects environmental data inside and outside vehicle, thus the pedestrian or vehicle to vehicle-surroundings carry out detection with
Identification.In this way when vehicle front is in the presence of to influence when vehicle is normally advanced or deviation occurs, and
When driver attention does not concentrate, vehicle-mounted ADAS device can issue information warning, the safety of support vehicles to vehicle driver
Traveling.
Specifically, in order to realize that vehicle-mounted ADAS device determines the identity legitimacy of driver, vehicle-mounted ADAS device in addition to
It needs to obtain outside the scene photograph of driver, it is also necessary to and cloud manages platform connection communication, to receive driver's network registry
When certificate photograph.
Specifically, vehicle-mounted ADAS device is provided with 4G module, can pass through 4G in order to manage platform connection communication with cloud
Module realizes the data interaction with cloud management platform.After driver gets on the bus, vehicle-mounted ADAS device can be by 4G module from cloud
End pipe platform downloads certificate photograph when driver's network registry, when driver identification identification authentication module identifies driver scene
When photo and cloud downloading photo are not same people, it is spare to do subsequent evidence obtaining that scene photograph can be passed back to cloud management platform.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations；To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement；And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (8)
1. a kind of low cost vision ADAS aims of systems detection recognition method, which is characterized in that including
Realize that the identification to driver authenticates；
The perception for realizing vehicle external environment, including the detection and identification to vehicle, pedestrian, lane line.
2. a kind of low cost vision ADAS aims of systems detects identification device characterized by comprising main equipment:
The main equipment includes: forward sight thermal infrared camera, backsight infrared camera, processor, alarm；
The forward sight thermal infrared camera, the backsight infrared camera and the alarm are connected to the processor；
The forward sight thermal infrared camera is used to obtain the driving video data of outside vehicle,
The backsight infrared camera is for obtaining driver's driving condition；
The processor is to receive video data and handle；
The processor is connected with cloud management platform, carries out data interaction.
3. according to claim 1, the 2 inexpensive vision ADAS aims of systems detection recognition method, which is characterized in that
The identification of the driver authenticate the following steps are included:
After driver gets on the bus, the backsight thermal infrared camera shoots driver's scene photograph；
Certificate photograph of the vehicle-mounted vision ADAS equipment from cloud management platform downloading driver login；
The identification authentication of driver is carried out according to the certificate photograph that scene photograph and cloud are downloaded.
4. the identification authentication according to claim 1, it is characterised in that:
System uses cascade one-stage deep learning network group, and the deep learning network group includes first network, the
Two networks and third network;
The first network is Face datection network;
Second network is face corrective network；
The third network is face feature coding network, can extract the facial characteristics vector of 1x512.
5. deep learning network group according to claim 4, it is characterised in that:
The first network is the one-stage detection network specially designed, is downloaded to the scene photograph and the cloud
Certificate photograph carries out Face datection, and segmentation extracts driver scene with a smaller size face picture and cloud downloading face
Picture；
Second network is the Face normalization network specially designed, can orient facial feature points, and face normalization is positive
Face;
The third network is the coding network specially designed, to download respectively to the live face picture and the cloud
Face picture extracts feature vector, and described eigenvector is the feature vector of 1x512.
6. deep learning network group according to claim 4, which is characterized in that
It in the network training process, needs according to sample feature, the features such as network task, net is detected to the one-stage
The loss function of network group is specially designed.
7. low cost vision ADAS aims of systems according to claim 2 detects identification device, which is characterized in that
In vehicle-mounted embedded type terminal, convolution is separated using depth and replaces traditional convolution, constructs lightweight deep neural network,
Reduce parameter amount and calculation amount；
The processor is combined convolution sum residual error network is separated, more efficient to the extraction of target signature；
The processor is connected with cloud management platform, carries out data interaction.
8. low cost vision ADAS aims of systems according to claim 2 detects identification device, including,
Driver identity identifies authentication module, for judging whether net the practical driver for joining the commerial vehicles such as vehicle, shared automobile
It is same people with registered drivers.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910346928.8A CN110110632A (en) | 2019-04-27 | 2019-04-27 | A kind of the vision ADAS target detection identifying system and device of low cost |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910346928.8A CN110110632A (en) | 2019-04-27 | 2019-04-27 | A kind of the vision ADAS target detection identifying system and device of low cost |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110110632A true CN110110632A (en) | 2019-08-09 |
Family
ID=67487061
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910346928.8A Pending CN110110632A (en) | 2019-04-27 | 2019-04-27 | A kind of the vision ADAS target detection identifying system and device of low cost |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110110632A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113162809A (en) * | 2021-04-30 | 2021-07-23 | 腾讯科技（深圳）有限公司 | Driving assistance processing method and device, computer readable medium and electronic device |
-
2019
- 2019-04-27 CN CN201910346928.8A patent/CN110110632A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113162809A (en) * | 2021-04-30 | 2021-07-23 | 腾讯科技（深圳）有限公司 | Driving assistance processing method and device, computer readable medium and electronic device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11488392B2 (en) | Vehicle system and method for detecting objects and object distance | |
| CN107967806B (en) | Vehicle fake-license detection method, device, readable storage medium storing program for executing and electronic equipment | |
| CN108513674B (en) | Detection and alarm method for accumulated snow and icing in front of vehicle, storage medium and server | |
| Nienaber et al. | Detecting potholes using simple image processing techniques and real-world footage | |
| CN109460699B (en) | Driver safety belt wearing identification method based on deep learning | |
| US9542609B2 (en) | Automatic training of a parked vehicle detector for large deployment | |
| CN111508260A (en) | Vehicle parking space detection method, device and system | |
| WO2019046400A1 (en) | Analyzing images and videos of damaged vehicles to determine damaged vehicle parts and vehicle asymmetries | |
| CN109583415A (en) | A kind of traffic lights detection and recognition methods merged based on laser radar with video camera | |
| CN108701396B (en) | Detection and alarm method for accumulated snow and icing in front of vehicle, storage medium and server | |
| CN103824452A (en) | Lightweight peccancy parking detection device based on full view vision | |
| CN110415544B (en) | Disaster weather early warning method and automobile AR-HUD system | |
| US11371851B2 (en) | Method and system for determining landmarks in an environment of a vehicle | |
| CN110658539B (en) | Vehicle positioning method, device, vehicle and computer readable storage medium | |
| CN110033622A (en) | Violation snap-shooting based on unmanned plane aerial photography technology occupies Emergency Vehicle Lane method | |
| CN111523352A (en) | Method for intelligently and rapidly identifying illegal modified vehicle and monitoring system thereof | |
| CN108509900A (en) | Road traffic sign detection and recognition methods, system and storage medium | |
| CN112793567A (en) | Driving assistance method and system based on road condition detection | |
| CN111967396A (en) | Processing method, device and equipment for obstacle detection and storage medium | |
| CN116829919A (en) | Automatic driving system testing method, testing system and storage medium based on aerial survey data | |
| US20220245951A1 (en) | Vehicle object detection | |
| Nambi et al. | ALT: towards automating driver license testing using smartphones | |
| CN110110632A (en) | A kind of the vision ADAS target detection identifying system and device of low cost | |
| CN110727269B (en) | Vehicle control method and related product | |
| CN112509321A (en) | Unmanned aerial vehicle-based driving control method and system for urban complex traffic situation and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190809 |
|
| WD01 | Invention patent application deemed withdrawn after publication |