CN108615028A - The fine granularity detection recognition method of harbour heavy vehicle - Google Patents

The fine granularity detection recognition method of harbour heavy vehicle Download PDF

Info

Publication number
CN108615028A
CN108615028A CN201810455259.3A CN201810455259A CN108615028A CN 108615028 A CN108615028 A CN 108615028A CN 201810455259 A CN201810455259 A CN 201810455259A CN 108615028 A CN108615028 A CN 108615028A
Authority
CN
China
Prior art keywords
vehicle
detection
camera
filter
window
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
Application number
CN201810455259.3A
Other languages
Chinese (zh)
Inventor
刘子健
史小林
刘鹤云
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Mainline Technology Co Ltd
Original Assignee
Beijing Mainline Technology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing Mainline Technology Co Ltd filed Critical Beijing Mainline Technology Co Ltd
Priority to CN201810455259.3A priority Critical patent/CN108615028A/en
Publication of CN108615028A publication Critical patent/CN108615028A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30248—Vehicle exterior or interior
    • G06T2207/30252—Vehicle exterior; Vicinity of vehicle
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/08—Detecting or categorising vehicles

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)
  • Traffic Control Systems (AREA)

Abstract

The present invention relates to a kind of fine granularity detection recognition methods of harbour heavy vehicle, camera lens is installed on driving vehicle towards the high definition camera in front of vehicle traveling, high definition camera is connect with computing module, the imaging data of camera is read using computing module, the vehicle detection identification within the scope of field of front vision is carried out, steps are as follows for specific method：1) image-forming information of acquisition target port vehicle in the camera；2) response transform and merging response；3) the fine granularity detection of vehicle；4) influence that different vehicle posture brings detection is solved.Advantageous effect：The present invention is by high definition camera and computing module, for the harbour heavy vehicle difference imaging results under all kinds of scenes, can make fast and accurately detection identification.

Description

The fine granularity detection recognition method of harbour heavy vehicle
Technical field
The invention belongs to field of vehicle detection more particularly to a kind of fine granularity detection recognition methods of harbour heavy vehicle.
Background technology
Harbour is important land and water distribution centre, and harbour heavy vehicle assumes responsibility for what large cargo was converted between transportation by land and water Key player.Traffic density is high under harbour service environment, and compared with middle-size and small-size vehicle, the driving difficulty of heavy vehicle is more Greatly, this just to driver, more stringent requirements are proposed.Therefore, research and development are put on for the auxiliary driving technology of harbour heavy vehicle Schedule, and wherein for the detection of harbour heavy vehicle and identification it is basis and the core of the technology.The main vehicle class at harbour It Wei not the uncommon vehicle of real roads such as truck (including tractor and vehicle hang), fork truck.Under harbour service environment, different appearances There are larger differences for the imaging of vehicle in the camera under state, and harbour truck has the short vehicle of carrying and hangs, long vehicle extension, loaded collection Vanning, unloaded container, all kinds of different state scene compositions such as loading container.
Current vehicle detection identifying schemes some by means of expensive laser radar apparatus, though however laser radar Barrier can so be effectively detected out, but its collected information density is small, it is highly difficult to extract target signature, therefore only It is difficult to identify vehicle in the environment by laser radar.It is some other using camera as the vehicle detection identification side of sensor Case rarely has for harbour service environment, and it is even more impossible to make fine-grained detection to harbour heavy vehicle to identify, system is driven for auxiliary The information for Decision Control provided of uniting is extremely limited.
Invention content
It is an object of the invention to overcome the defect of prior art, and provide a kind of detections of the fine granularity of harbour heavy vehicle to identify Method carries out the fine-grained detection identification of vehicle, significantly carries for the harbour heavy vehicle difference imaging results under all kinds of scenes The high detection recognition accuracy of harbour heavy vehicle.
The present invention to achieve the above object, is achieved through the following technical solutions, a kind of fine granularity inspection of harbour heavy vehicle Detection identifying method, it is characterized in that：Installation camera lens is towards the high definition camera in front of vehicle traveling on driving vehicle, by high definition camera It is connect with computing module, the imaging data of camera is read using computing module, the vehicle detection carried out within the scope of field of front vision is known Not, steps are as follows for specific method：
1) image-forming information of acquisition target port vehicle in the camera
1. acquiring harbour service ambient image data, and the vehicle detection target in image is labeled, will be marked Image sample data be used for model training, build multigroup filter template and the deformable component feature space model of target, extract Spatial model and filter template；
2. extracting candidate window on the image marked using sliding window method, the direction gradient histogram of image is calculated Figure feature, according to the global root filter of histograms of oriented gradients pyramid training and multiple partial models, wherein each component Model all includes a component filter and a spatial model；The component filter for respond headstock, vehicle extension, wheel, Container class vehicle assembly, spatial model define the relative position that each component detected occurs in global detection window, The deformation loss brought for calculating unit change in location；
2) response transform and merging response
1. in detection process, for the image collected data, extracting its histograms of oriented gradients feature pyramid, calculate Response between low resolution feature and root filter；
2. calculating the response between high-resolution features and all parts filter, size change over is done to response computation result Afterwards, merge two-part response computation as a result, making comparisons with the spatial model that training obtains before, calculate setting for each window x Confidence score, calculation formula are：
fβ=maxzβ Φ (x, z)
Wherein：
Z is the relative position that each component occurs in the window
β is filter and spatial model parameter；
3. a threshold value is arranged to the confidence level of window, the window that score is more than to threshold value is set as high credible of fitting degree Testing result；
3) the fine granularity detection of vehicle
Different filter templates is built for the heavy vehicle of different scenes, each group that the same window uses is filtered Device calculates response, and finding makes the highest filter group of window scores, and the corresponding scene of this group of filter is testing result, Realize fine-grained detection identification；
4) influence that different vehicle posture brings detection is solved
It is imaged existing difference in the camera for vehicle difference posture, with the method for structure spatial model, vehicle Different postures will cause the relative position that each component of vehicle occurs in detection window different, and spatial model contains vehicle each group The various positions information that part is likely to occur in detection window can be fitted the position distribution situation of each unit response, and solution is never The influence that detection is brought with vehicle attitude.
The high definition camera uses frame rate in 20fps or more, the imaging color camera more than 720P resolution ratio.
The computing module selects computing modules of the CPU more than dominant frequency 2.0GHz.It adjusts camera focus and is not less than 30 meters.
Advantageous effect：Compared with prior art, the present invention is by slr camera and computing module, independent of laser radar Or other sensors, it is simple in structure, convenient for transplanting.For the harbour heavy vehicle difference imaging results under all kinds of scenes, equal energy Fast and accurately detection identification is enough made, speed is not less than 20fps, the detection within the scope of 20 meters of camera visible angle to vehicle Rate is not less than 90%.
Description of the drawings
Fig. 1 is the work block diagram of the present invention.
Fig. 2 is the scheme of installation of apparatus of the present invention.
A, camera model, B, computing module, C, camera fields of view, D, car body.
Specific implementation mode
Below in conjunction with preferred embodiment, to the specific implementation mode that provides according to the present invention, details are as follows：
Attached drawing 1 is referred to, present embodiment discloses a kind of fine granularity detection recognition methods of harbour heavy vehicle, are driving vehicle High definition camera connect with computing module towards the high definition camera in vehicle traveling front, uses computing module by installation camera lens on The imaging data of camera is read, carries out the vehicle detection identification within the scope of field of front vision, steps are as follows for specific method：
1) image-forming information of acquisition target port vehicle in the camera
1. acquiring harbour service ambient image data, and the vehicle detection target in image is labeled, will be marked Image sample data be used for model training, build multigroup filter template and the deformable component feature space model of target, extract Spatial model and filter template；
2. extracting candidate window on the image marked using sliding window method, the direction gradient histogram of image is calculated Figure feature, according to the global root filter of histograms of oriented gradients pyramid training and multiple partial models, wherein each component Model all includes a component filter and a spatial model；The component filter for respond headstock, vehicle extension, wheel, Container class vehicle assembly, spatial model define the relative position that each component detected occurs in global detection window, The deformation loss brought for calculating unit change in location；
2) response transform and merging response
1. in detection process, for the image collected data, extracting its histograms of oriented gradients feature pyramid, calculate Response between low resolution feature and root filter；
2. calculating the response between high-resolution features and all parts filter, size change over is done to response computation result Afterwards, merge two-part response computation as a result, making comparisons with the spatial model that training obtains before, calculate setting for each window x Confidence score, calculation formula are：
fβ=maxzβ Φ (x, z)
Wherein：
Z is the relative position that each component occurs in the window
β is filter and spatial model parameter；
3. a threshold value is arranged to the confidence level of window, the window that score is more than to threshold value is set as high credible of fitting degree Testing result；
3) the fine granularity detection of vehicle
Different filter templates is built for the heavy vehicle of different scenes, each group that the same window uses is filtered Device calculates response, and finding makes the highest filter group of window scores, and the corresponding scene of this group of filter is testing result, Realize fine-grained detection identification；
4) influence that different vehicle posture brings detection is solved
It is imaged existing difference in the camera for vehicle difference posture, with the method for structure spatial model, vehicle Different postures will cause the relative position that each component of vehicle occurs in detection window different, and spatial model contains vehicle each group The various positions information that part is likely to occur in detection window can be fitted the position distribution situation of each unit response, and solution is never The influence that detection is brought with vehicle attitude.
The high definition camera uses frame rate in 20fps or more, the imaging color camera more than 720P resolution ratio.
The computing module selects computing modules of the CPU more than dominant frequency 2.0GHz.It adjusts camera focus and is not less than 30 meters.
Embodiment
Detection device is installed as shown in Figure 2, and camera model A is mounted at the top of car body D, and frame rate 20fps is differentiated Rate is 720P, towards vehicle heading, adjusts camera fields of view C, and it is 30 meters to make its focal length, and is by itself and CPU frequency The computing module B connections of 2.0GHz.Image data is acquired using camera, the image sample data marked is instructed for model Practice.During model training, setting model learning rate is 0.001, and model parameter is carried out more using stochastic gradient descent method Newly, loss function uses log-likelihood function, and adds L2 regular terms, and regularization coefficient is set as 0.05.It is carried after the completion of training Take out model.After collecting the image data for detection, model is entered data into, model will export the result of vehicle detection.
Reliability of this method for vehicle detection identification is assessed using accuracy rate and recall rate.In front of high definition camera Within the scope of 30 meters of the visual field, this method can effectively detect that 90% or more vehicle, probability of false detection are no more than 30%, that is, are ensureing In the case that recall rate is not less than 90%, Detection accuracy is not less than 70%, can be provided with robust for automatic Pilot technology The vehicle cognitive method of property, valuable decision information is provided for the decision process of automatic Pilot.
In implementation process, need to solve the influence that different vehicle posture brings detection.Since the different postures of vehicle will Cause the relative position that each component of vehicle occurs in detection window different, using the method for structure spatial model, spatial model The various positions information that each component of vehicle is likely to occur in detection window is contained, the position point of each unit response can be fitted Cloth situation, to solve the influence that different vehicle posture brings detection.
The above-mentioned detailed description that a kind of fine granularity detection recognition method of the harbour heavy vehicle is carried out with reference to embodiment, It is illustrative without being restrictive, several embodiments can be enumerated according to limited range, therefore do not departing from this hair Change and modification under bright general plotting should belong within protection scope of the present invention.

Claims (3)

1. a kind of fine granularity detection recognition method of harbour heavy vehicle, it is characterized in that：Camera lens direction is installed on driving vehicle The high definition camera in vehicle traveling front, high definition camera is connect with computing module, the imaging number of camera is read using computing module According to the vehicle detection carried out within the scope of field of front vision identifies that steps are as follows for specific method：
1) image-forming information of acquisition target port vehicle in the camera
1. acquiring harbour service ambient image data, and the vehicle detection target in image is labeled, the figure that will have been marked Decent notebook data is used for model training, builds multigroup filter template and the deformable component feature space model of target, extracts space Model and filter template；
2. extracting candidate window on the image marked using sliding window method, the histograms of oriented gradients for calculating image is special Sign, according to the global root filter of histograms of oriented gradients pyramid training and multiple partial models, wherein each partial model All include a component filter and a spatial model；The component filter is for responding headstock, vehicle extension, wheel, container Class vehicle assembly, spatial model define the relative position that each component detected occurs in global detection window, are used for The deformation loss that calculating unit change in location is brought；
2) response transform and merging response
1. in detection process, for the image collected data, extracting its histograms of oriented gradients feature pyramid, calculating low point Response between resolution feature and root filter；
2. the response between high-resolution features and all parts filter is calculated, after doing size change over to response computation result, Merge two-part response computation as a result, making comparisons with the spatial model that training obtains before, calculates the confidence level of each window x Score, calculation formula are：
fβ=maxzβ Φ (x, z)
Wherein：
Z is the relative position that each component occurs in the window
β is filter and spatial model parameter；
3. a threshold value is arranged in the confidence level to window, the window that score is more than to threshold value is set as the high credible detection of fitting degree As a result；
3) the fine granularity detection of vehicle
Different filter templates is built for the heavy vehicle of different scenes, each group filter meter used the same window Response is calculated, finding makes the highest filter group of window scores, and the corresponding scene of this group of filter is testing result, realizes Fine-grained detection identification；
4) influence that different vehicle posture brings detection is solved
It is imaged existing difference in the camera for vehicle difference posture, with the method for structure spatial model, the difference of vehicle Posture will cause the relative position that each component of vehicle occurs in detection window different, and spatial model contains each component of vehicle and exists The various positions information being likely to occur in detection window can be fitted the position distribution situation of each unit response, solve different vehicles The influence that posture brings detection.
2. the fine granularity detection recognition method of harbour heavy vehicle according to claim 1, it is characterized in that：The high definition phase Machine uses the frame rate of 20fps or more, imaging color camera more than 720P resolution ratio.
3. the fine granularity detection recognition method of harbour heavy vehicle according to claim 1, it is characterized in that：The calculating mould Block selects computing modules of the CPU more than dominant frequency 2.0GHz.It adjusts camera focus and is not less than 30 meters.
CN201810455259.3A 2018-05-14 2018-05-14 The fine granularity detection recognition method of harbour heavy vehicle Pending CN108615028A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810455259.3A CN108615028A (en) 2018-05-14 2018-05-14 The fine granularity detection recognition method of harbour heavy vehicle

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810455259.3A CN108615028A (en) 2018-05-14 2018-05-14 The fine granularity detection recognition method of harbour heavy vehicle

Publications (1)

Publication Number Publication Date
CN108615028A true CN108615028A (en) 2018-10-02

Family

ID=63663306

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810455259.3A Pending CN108615028A (en) 2018-05-14 2018-05-14 The fine granularity detection recognition method of harbour heavy vehicle

Country Status (1)

Country Link
CN (1) CN108615028A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109597097A (en) * 2018-12-06 2019-04-09 北京主线科技有限公司 Scan-type obstacle detection method based on multi-thread laser
CN111428730A (en) * 2019-01-09 2020-07-17 中国科学技术大学 Weak supervision fine-grained object classification method

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104636749A (en) * 2013-11-14 2015-05-20 中国移动通信集团公司 Target object detection method and device
CN104751190A (en) * 2015-04-23 2015-07-01 武汉大学 Vehicle part positioning method for vehicle fine recognition
CN106067022A (en) * 2016-05-28 2016-11-02 北方工业大学 Remote sensing image harbor ship detection false alarm eliminating method based on DPM algorithm
CN106384100A (en) * 2016-09-28 2017-02-08 武汉大学 Component-based fine vehicle model recognition method
CN106504580A (en) * 2016-12-07 2017-03-15 深圳市捷顺科技实业股份有限公司 A kind of method for detecting parking stalls and device
CN106600631A (en) * 2016-11-30 2017-04-26 郑州金惠计算机系统工程有限公司 Multiple target tracking-based passenger flow statistics method

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104636749A (en) * 2013-11-14 2015-05-20 中国移动通信集团公司 Target object detection method and device
CN104751190A (en) * 2015-04-23 2015-07-01 武汉大学 Vehicle part positioning method for vehicle fine recognition
CN106067022A (en) * 2016-05-28 2016-11-02 北方工业大学 Remote sensing image harbor ship detection false alarm eliminating method based on DPM algorithm
CN106384100A (en) * 2016-09-28 2017-02-08 武汉大学 Component-based fine vehicle model recognition method
CN106600631A (en) * 2016-11-30 2017-04-26 郑州金惠计算机系统工程有限公司 Multiple target tracking-based passenger flow statistics method
CN106504580A (en) * 2016-12-07 2017-03-15 深圳市捷顺科技实业股份有限公司 A kind of method for detecting parking stalls and device

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
PEDRO F. FELZENSZWALB ET AL.: "Object Detection with Discriminatively Trained Part-Based Models", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 *
康珮珮 等: "车辆检测中可变形部件模型的改进与应用", 《计算机工程与应用》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109597097A (en) * 2018-12-06 2019-04-09 北京主线科技有限公司 Scan-type obstacle detection method based on multi-thread laser
CN109597097B (en) * 2018-12-06 2023-04-18 北京主线科技有限公司 Scanning type obstacle detection method based on multi-line laser
CN111428730A (en) * 2019-01-09 2020-07-17 中国科学技术大学 Weak supervision fine-grained object classification method
CN111428730B (en) * 2019-01-09 2022-07-08 中国科学技术大学 Weak supervision fine-grained object classification method

Similar Documents

Publication Publication Date Title
CN107463890B (en) A kind of Foregut fermenters and tracking based on monocular forward sight camera
CN105373135B (en) A kind of method and system of aircraft docking guidance and plane type recognition based on machine vision
CN105835880B (en) Lane following system
CN104573646B (en) Chinese herbaceous peony pedestrian detection method and system based on laser radar and binocular camera
CN102956106B (en) For identifying that motor vehicles are to monitor the method and apparatus of traffic
US20200041284A1 (en) Map road marking and road quality collecting apparatus and method based on adas system
CN104616502B (en) Car license recognition and alignment system based on combination type bus or train route video network
CN110415544B (en) Disaster weather early warning method and automobile AR-HUD system
US9626599B2 (en) Reconfigurable clear path detection system
CN106096525A (en) A kind of compound lane recognition system and method
CN106127107A (en) The model recognizing method that multi-channel video information based on license board information and vehicle's contour merges
CN109711353B (en) Ship waterline area identification method based on machine vision
CN110658539B (en) Vehicle positioning method, device, vehicle and computer readable storage medium
CN107796373A (en) A kind of distance-finding method of the front vehicles monocular vision based on track plane geometry model-driven
CN107798688A (en) Motion estimate method, method for early warning and automobile anti-rear end collision prior-warning device
CN106778540A (en) Parking detection is accurately based on the parking event detecting method of background double layer
CN113034378A (en) Method for distinguishing electric automobile from fuel automobile
CN108615028A (en) The fine granularity detection recognition method of harbour heavy vehicle
CN111881984A (en) Target detection method and device based on deep learning
CN108256418B (en) Pedestrian early warning method and system based on infrared imaging
CN111414857B (en) Front vehicle detection method based on vision multi-feature fusion
CN108198428A (en) Lorry intercepting system and hold-up interception method
CN111539278A (en) Detection method and system for target vehicle
CN116520351A (en) Train state monitoring method, system, storage medium and terminal
CN115857040A (en) Dynamic visual detection device and method for foreign matters on locomotive roof

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
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20181002

WD01 Invention patent application deemed withdrawn after publication