WO2018188309A1 - 行人识别装置及方法、辅助驾驶装置 - Google Patents

行人识别装置及方法、辅助驾驶装置 Download PDF

Info

Publication number
WO2018188309A1
WO2018188309A1 PCT/CN2017/107434 CN2017107434W WO2018188309A1 WO 2018188309 A1 WO2018188309 A1 WO 2018188309A1 CN 2017107434 W CN2017107434 W CN 2017107434W WO 2018188309 A1 WO2018188309 A1 WO 2018188309A1
Authority
WO
WIPO (PCT)
Prior art keywords
particle
pedestrian
image
sub
circuit
Prior art date
Application number
PCT/CN2017/107434
Other languages
English (en)
French (fr)
Inventor
苏海军
唐小军
Original Assignee
京东方科技集团股份有限公司
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 京东方科技集团股份有限公司 filed Critical 京东方科技集团股份有限公司
Priority to EP17882265.6A priority Critical patent/EP3611652A4/en
Priority to US16/066,799 priority patent/US11508157B2/en
Publication of WO2018188309A1 publication Critical patent/WO2018188309A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W30/00—Purposes of road vehicle drive control systems not related to the control of a particular sub-unit, e.g. of systems using conjoint control of vehicle sub-units
    • B60W30/08—Active safety systems predicting or avoiding probable or impending collision or attempting to minimise its consequences
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/211—Selection of the most significant subset of features
    • G06F18/2111—Selection of the most significant subset of features by using evolutionary computational techniques, e.g. genetic algorithms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/771—Feature selection, e.g. selecting representative features from a multi-dimensional feature space
    • 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
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B60W2050/0001—Details of the control system
    • B60W2050/0019—Control system elements or transfer functions

Definitions

  • the invention belongs to the technical field of pedestrian recognition, and in particular relates to a pedestrian recognition device and method and an auxiliary driving device.
  • Pedestrian identification is a key technology for assisting driving devices or intelligent monitoring devices. It mainly uses image processing technology to identify pedestrians in the target area in real time, providing technical basis for real-time alarms.
  • the commonly used pedestrian recognition device adopts a camera and a computer vision technology.
  • the hardware device implementing the method is simple and the cost is low, but the software algorithm for implementing the method is complicated. How to improve recognition accuracy and recognition rate is often the focus of attention in this field.
  • the recognition rate is more important than the recognition accuracy.
  • the commonly used pedestrian recognition device uses a sliding window to traverse the search image, and the recognition rate is low, which cannot meet the requirements.
  • the present invention aims to at least solve one of the technical problems existing in the prior art, and proposes a pedestrian recognition device and method and an auxiliary driving device. According to the present disclosure, not only the recognition rate but also different recognition rates can be satisfied. Application scenarios for requirements.
  • an embodiment of the present disclosure provides a pedestrian recognition apparatus, including: a classifier training circuit, an image acquisition circuit, a sampling circuit, and a calculation circuit;
  • the classifier training circuit is configured to extract a pedestrian feature based on the training sample, and perform offline training based on the pedestrian feature to obtain a classifier;
  • the image acquisition circuit is configured to collect an image to be identified
  • the sampling circuit is configured to sample a sub-image on the image to be identified
  • the calculation circuit is configured to identify in the sub-image based on a particle swarm optimization algorithm a pedestrian, wherein each particle is defined as an object of a preset size in the sub-image, and a fitness value of the particle is calculated based on the classifier and a pedestrian feature of each particle in a particle swarm optimization algorithm, The fitness value characterizes the likelihood that a particle belongs to a pedestrian.
  • the pedestrian feature comprises an integral channel feature.
  • the classifier includes a boost classifier.
  • the calculation circuit comprises:
  • Initializing the sub-circuit configured to initialize the maximum number of iterations, the number of particles, the position of each particle, the parameters in the velocity update formula, and the parameters in the position update formula;
  • a pedestrian feature extraction sub-circuit configured to extract a pedestrian feature of each particle
  • the fitness value calculation sub-circuit is configured to calculate a fitness value of the particle based on the classifier and a pedestrian characteristic of each particle;
  • the optimal position determining sub-circuit is configured to compare the fitness value of each particle with the fitness value of the search position of the particle, and determine the maximum value as the local optimal solution of the particle in the search process, and in all particles In the local optimal solution, the maximum value of the fitness value is determined as the global optimal solution;
  • a particle velocity update subcircuit configured to update a velocity of each particle based on a global optimal solution and a local optimal solution
  • a particle position update sub-circuit configured to update a position of each particle based on the updated particle velocity to obtain a next generation particle
  • a result output sub-circuit configured to determine whether the fitness value of each of the next generation particles calculated by the fitness value calculation sub-circuit reaches a preset value, wherein, if so, the pedestrian recognition device determines that the particle belongs to a pedestrian; No, the pedestrian identification device controls the optimal position determining sub-circuit to operate;
  • the iteration count statistic sub-circuit is configured to add 1 to the current iteration number after obtaining the next generation of particles, and send a stop work instruction to the pedestrian feature extraction sub-circuit when the number of iterations reaches the maximum number of iterations.
  • the sampling circuit is configured to perform stepwise scaling on the image to be recognized according to a preset scaling factor to obtain a multi-level sub-image
  • the calculation circuit is configured to identify a pedestrian based on a particle swarm optimization algorithm for a next-level sub-image when the number of iterations reaches a maximum number of iterations.
  • An embodiment of the present disclosure also provides a pedestrian recognition method, including the following steps:
  • the pedestrian feature calculates a fitness value for the particle, the fitness value indicating the likelihood that the particle belongs to a pedestrian.
  • the pedestrian feature comprises an integral channel feature.
  • the classifier includes a boost classifier.
  • the step of identifying a pedestrian based on the particle swarm optimization algorithm for the sub-image comprises:
  • the initialization step includes initializing the maximum number of iterations, the number of particles, the position of each particle, the parameters in the velocity update formula, and the parameters in the position update formula;
  • the number of current iterations is increased by one after the next generation of particles is obtained, and the work step is stopped when the number of iterations reaches the maximum number of iterations.
  • the step of obtaining a sub-image on the image to be identified includes:
  • the pedestrian recognition method further includes: identifying the pedestrian based on the particle swarm optimization algorithm for the next-level sub-image when the number of iterations reaches the maximum number of iterations.
  • An embodiment of the present disclosure also provides an assisting driving device including the pedestrian recognition device provided by the present invention.
  • an embodiment of the present invention further provides an electronic device including: a housing, a processor, a memory, a circuit board, and a power supply circuit, wherein the circuit board is disposed inside a space enclosed by the housing, the processor and the memory Provided on a circuit board; a power supply circuit for supplying power to each circuit or device of the above electronic device; and a memory for storing executable program code, and
  • the processor executes the following steps by reading executable program code stored in the memory and running a program corresponding to the executable program code: extracting a pedestrian feature based on the training sample, and obtaining a classifier based on the pedestrian feature offline training;
  • the pedestrian feature calculates a fitness value for the particle, the fitness value indicating the likelihood that the particle belongs to a pedestrian.
  • FIG. 1 is a schematic block diagram of a pedestrian recognition apparatus according to an embodiment of the present invention.
  • FIG. 2 is a schematic block diagram of the calculation circuit of FIG. 1;
  • FIG. 3 is a flowchart of a pedestrian recognition method according to an embodiment of the present invention.
  • step S40 of FIG. 4 is a flow chart of step S40 of FIG.
  • FIG. 1 is a schematic block diagram of a pedestrian recognition apparatus according to an embodiment of the present invention.
  • a pedestrian recognition apparatus according to an embodiment of the present invention includes a classifier training circuit 10, an image acquisition circuit 20, a sampling circuit 30, and a calculation circuit 40.
  • the classifier training circuit 10 is configured to extract a pedestrian feature based on the training sample, and perform training based on the pedestrian feature offline to obtain a classifier.
  • the image acquisition circuit 20 is configured to acquire an image to be identified.
  • the sampling circuit 30 is configured to sample on the image to be identified to obtain a sub-image.
  • the calculation circuit 40 is configured to identify pedestrians in the sub-image based on the particle swarm optimization algorithm, wherein each particle is defined as an object of a preset size in the sub-image; based on the classifier and the pedestrian characteristics of each particle in the particle swarm optimization algorithm
  • the fitness value of the particle is calculated, and the fitness value characterizes the probability that the particle belongs to the pedestrian. Specifically, if the fitness value is larger, the probability of belonging to the pedestrian is also large; if the fitness value is smaller, the possibility of belonging to the pedestrian is smaller.
  • the image pickup circuit 20 may be formed as a camera or the like having an imaging function
  • the calculation circuit 40 may be formed as a processor or the like.
  • pedestrian features include, but are not limited to, integral channel features. Since the integration channel feature can better describe pedestrians, it is possible to improve the accuracy and recognition efficiency of identifying pedestrians.
  • the pedestrian recognition apparatus calculates a feature including a LUV (chroma and chromatic aberration), a gradient magnitude, and a direction gradient histogram (HOG), and It is connected in sequence to form an integral channel feature, wherein LUV represents brightness and two color difference characteristics respectively; gradient amplitude is the square root of adjacent pixel difference of up, down, left and right, and gradient direction is the inverse tangent of adjacent pixel difference of up, down, left and right (formula) (1)), mag represents the gradient magnitude, Ori represents the gradient direction; HOG is the weighted histogram representation of the gradient magnitude of each pixel in the corresponding gradient direction (formula (2)):
  • the classifier includes, but is not limited to, a boost classifier.
  • the boost classifier is the process of hardening several weak classifiers into one strong classifier. Therefore, the boost classifier has the characteristics of flexible use and high classification accuracy, and also takes into account the calculation performance.
  • the calculation circuit 40 includes an initialization sub-circuit 401, a pedestrian feature extraction sub-circuit 402, a fitness calculation sub-circuit 403, an optimal position determination sub-circuit 404, a particle velocity update sub-circuit 405, and a particle position.
  • the sub-circuit 406, the result output sub-circuit 407, and the iteration count statistic sub-circuit 408 are updated.
  • the initialization module 401 is configured to initialize the maximum number of iterations, the number of particles, the position of each particle, the parameters in the velocity update formula, and the parameters in the location update formula.
  • the maximum number of iterations and the number of particles are usually set by the user; the position of each particle initialized is usually set to a random value; the parameters in the speed update formula and the position update formula can be obtained empirically.
  • Optimal constant value For pedestrian recognition, the search space is a two-dimensional space. Therefore, the position of the particle can be represented by (x i1 , x i2 ). Each particle has not only a position attribute, but also a speed attribute and a fitness value attribute.
  • the pedestrian feature extraction sub-circuit 402 is configured to extract the pedestrian features of each particle.
  • the fitness value calculation sub-circuit 403 is configured to calculate the fitness value of the particle based on the classifier and the pedestrian characteristics of each particle. Specifically, fitness values are used to measure the pros and cons of particles, and also serve as the basis for determining local and global optimal particles, and are the basis for particle search.
  • the optimal position determining sub-circuit 404 is configured to compare the fitness value of each particle with the fitness value of the previous search position of the particle, and determine the maximum value as the local optimal solution of the particle during the search process, and in all the particles In the local optimal solution, the maximum value of the fitness value is determined as the global optimal solution.
  • the particle velocity update sub-circuit 405 is configured to update the velocity of each particle based on the global optimal solution and the local optimal solution. Specifically, since the motion of all particles is affected by the local optimal solution and the global optimal solution, the speed update formula includes but is not limited to the following:
  • ⁇ is the coefficient that maintains the original speed, so it is called inertia weight, usually set to 0.79;
  • c 1 is the weight coefficient of the particle tracking its own historical optimal value, which represents the cognition of the particle itself, so it is called the cognitive coefficient, usually set to 2;
  • c 2 is the weight coefficient of the particle tracking population optimal value, which represents the particle's knowledge of the entire group knowledge, so it is called the social coefficient, usually set to 2;
  • ⁇ , ⁇ is a uniformly distributed random number in the interval [0,1], the role of which is to generate random perturbations, usually set to 0.7;
  • d represents the dimension, and there are only two dimensions of x and y for pedestrian recognition
  • the particle position update sub-circuit 406 is configured to update the position of each particle based on the updated particle velocity to obtain next generation particles. Specifically, since each particle has the velocity calculated above, the position of the particle is updated by the following formula driven by the speed:
  • r is the speed constraint and is usually set to 1.
  • the particles are always kept within the search space.
  • the particle size is 64*128, so the search space is: [m-64, n-128].
  • the result output sub-circuit 407 is configured to determine whether the fitness value of each of the next generation particles calculated by the fitness value calculation sub-circuit 403 reaches a preset value, wherein, if so, the line
  • the person identification device determines that the particle belongs to a pedestrian; if not, the pedestrian identification device controls the optimal position determining sub-circuit 404 to continue to operate. It can be understood that the recognition rate can be further improved by using the preset value as a threshold to judge whether the particles belong to a pedestrian.
  • the resulting output sub-circuit determines that the particle belongs to a pedestrian, the image corresponding to the particle is superimposed in the output image, and/or an alarm is provided.
  • the iteration count statistic sub-circuit 408 is configured to cumulatively increment the number of iterations after obtaining the next generation of particles, and to send a stop work instruction to the pedestrian feature extraction sub-circuit 402 when the number of iterations reaches the maximum number of iterations.
  • the sampling circuit 30 is configured to scale the image to be recognized according to a preset scaling factor to obtain a multi-level sub-image; and the calculation circuit 40 is configured to be based on the next-level sub-image when the number of iterations reaches the maximum number of iterations
  • the particle swarm optimization algorithm identifies pedestrians.
  • the image to be recognized by the sampling circuit 30 is stepwise scaled according to a preset scaling factor to obtain a multi-level sub-image, so that pedestrians can be identified in the enlarged sub-image when the pedestrian is small, and the pedestrian is larger in the sub-image.
  • the pedestrian recognition apparatus according to the embodiment of the present invention can quickly recognize a pedestrian.
  • the sampling formula can be expressed as: scale * [m, n], where scale is a scaling factor. If the scaling factor is greater than 1, the sampled image width and height of each level of the sample will be scaled up relative to the previous level of the sub-image. At this point, the sub-image can be said to be upsampled. If 0 ⁇ scale ⁇ 1, the sub-image width and height of each sample sampled will be scaled down relative to the previous sub-image. At this point, the sub-image can be referred to as downsampling.
  • the obtained multi-level sub-images may be referred to as a first-level sub-image, a second-level sub-image, etc., wherein the first-level sub-image is a previous-level sub-image of the second-level sub-image.
  • the working process of the pedestrian recognition apparatus provided by the embodiment of the present invention is described in detail below. Specifically, the working process includes the following steps:
  • the initialization sub-circuit 401 initializes the maximum number of iterations, the number of particles, the position of each particle, the parameter in the velocity update formula, and the parameter in the position update formula;
  • the sampling circuit 30 performs stepwise scaling on the image to be recognized according to a preset scaling factor to obtain a multi-level sub-image.
  • the working process also performs the following steps:
  • the pedestrian feature extraction sub-circuit 402 extracts the pedestrian characteristics of each particle
  • the fitness value calculation sub-circuit 403 calculates the fitness value of the particle based on the classifier and the pedestrian characteristics of each particle;
  • the optimal position determining sub-circuit 404 determines a local optimal solution and a global optimal solution
  • the particle velocity update sub-circuit 405 updates the velocity of each particle based on the local optimal solution and the global optimal solution;
  • the particle position update sub-circuit 406 updates the position of each particle based on the updated particle velocity to obtain next generation particles.
  • the pedestrian feature extraction sub-circuit 402 extracts pedestrian characteristics of the next generation particle
  • the fitness value calculation sub-circuit 403 calculates the fitness value of the particle based on the classifier and the pedestrian characteristics of each of the next generation particles;
  • the result output sub-circuit 407 determines whether the fitness value of each of the next generation particles reaches a preset value. If yes, the pedestrian recognition device determines that the particle belongs to a pedestrian, and superimposes the image corresponding to the particle in the output image. And/or providing an alarm; if not, the pedestrian identification device performs step S12 and then returns to step S6;
  • the iteration count statistic sub-circuit 408 adds 1 to the current number of iterations after obtaining the next generation of particles, and performs step S4 on the next-level sub-image when the number of iterations reaches the maximum number of iterations.
  • step S10 If no pedestrian is recognized in steps S10 and S11 for the multi-level sub-image, the process proceeds to step S2, and the next frame image of the video is taken as the image to be recognized, and then continues to operate.
  • the pedestrian recognition apparatus refers to the field of pedestrian recognition by using a particle swarm optimization algorithm, and generalizes a particle into a pedestrian recognition window with the size of the particle as a starting point, and adapts The degree value is generalized to characterize the probability scores belonging to pedestrians.
  • Particle Swarm Optimization (PSO) algorithm can quickly converge to the region with higher probability score in the search image, and can make the recognition rate of pedestrian recognition no longer proportional to the image size, but with the optimization algorithm.
  • the number of particles is correlated so that it can meet the application scenarios of different recognition rate requirements.
  • FIG. 3 is a flowchart of a pedestrian recognition method according to an embodiment of the present invention.
  • the pedestrian recognition method provided by the embodiment of the present invention includes the following steps:
  • each particle is defined as an object of a preset size in the sub-image, and based on the classifier and each particle in the particle swarm optimization algorithm
  • the pedestrian feature calculates the fitness value of the particle, and the fitness value indicates the probability that the particle belongs to the pedestrian.
  • the pedestrian feature includes an integral channel feature.
  • the classifier includes a boost classifier.
  • step S40 includes the following steps:
  • An initialization step including initializing a maximum number of iterations, a number of particles, a position of each particle, a parameter in a speed update formula, and a parameter in a position update formula;
  • the current iteration number is incremented by 1 after the next generation of particles is obtained, and the work step is stopped when the number of iterations reaches the maximum number of iterations.
  • step S30 includes: treating the image to be recognized according to a preset scaling factor The lines are scaled step by step to obtain multiple sub-images.
  • the step S4022 further includes: identifying the pedestrian based on the particle swarm optimization algorithm for the next-level sub-image.
  • the pedestrian recognition method provided by the embodiment of the present invention corresponds to the pedestrian recognition device provided by the above embodiment of the present invention, and the pedestrian recognition device has been described in detail above. Therefore, the pedestrian recognition method provided by the embodiment is related.
  • the content will not be described in detail here, please refer to the corresponding content in the pedestrian identification device.
  • the particle swarm optimization algorithm refers to the field of pedestrian recognition, and generalizes a particle into a pedestrian recognition window with the size of the particle as a starting point, and The fitness value is generalized to characterize the probability scores belonging to pedestrians.
  • Particle swarm optimization (PSO) algorithm can quickly converge to the region with higher probability score in the search image, and can make the recognition rate of pedestrian recognition no longer proportional to the image size, but is related to the particle number of the optimization algorithm. So that it can meet the application scenarios of different recognition rate requirements.
  • An embodiment of the present invention also provides an assisting driving device comprising the pedestrian recognition device provided by the first embodiment described above.
  • the assisting driving device may include a camera, a processor, a memory, and the like.
  • the processor can be integrated with a navigation system or a zoom system in a structure such as a central control panel, a rear view mirror, or a driving recording device.
  • the assist driving device further includes an auxiliary driver driving device.
  • the assist driving device further includes an auxiliary device in a driverless car or the like.
  • the pedestrian recognition device provided by the above-described first embodiment of the present invention since the pedestrian can be quickly recognized. Therefore, a good assist driving effect can be obtained, and the applicability is strong.
  • An embodiment of the present invention further provides an electronic device comprising: a housing, a processor, a memory, a circuit board, and a power supply circuit, wherein the circuit board is disposed inside the space enclosed by the housing, and the processor and the memory are disposed in the circuit On the board; a power circuit for powering various circuits or devices of the above electronic device; and a memory for storing executable program code.
  • the processor performs the following steps by reading executable program code stored in the memory and executing a program corresponding to the executable program code: extracting pedestrian features based on the training samples, and based on The pedestrian feature offline training obtains a classifier;
  • the pedestrian feature calculates a fitness value for the particle, the fitness value indicating the likelihood that the particle belongs to a pedestrian.
  • the electronic device may be integrated in a structure such as a central control panel, a rear view mirror or a driving recording device of the vehicle, or may be independent of a central control panel, a rear view mirror or a driving recording device of the vehicle.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Human Computer Interaction (AREA)
  • Medical Informatics (AREA)
  • Automation & Control Theory (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Mechanical Engineering (AREA)
  • Transportation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Physiology (AREA)
  • Image Analysis (AREA)
  • Traffic Control Systems (AREA)

Abstract

提供了一种行人识别装置，包括：分类器训练电路（10），构造成基于训练样本提取行人特征，并基于该行人特征离线训练获得分类器；图像采集电路（20），构造成采集待识别图像；采样电路（30），构造成在待识别图像上采样获得子图像；计算电路（40），构造成对子图像基于粒子群优化算法识别行人，其中，每个粒子定义为子图像中预设大小的对象，并且在粒子群优化算法中基于分类器和每个粒子的行人特征计算该粒子的适应度值，适应度值表征粒子属于行人的可能性大小。还提供了一种行人识别方法和一种驾驶辅助装置。上述方法和装置，不仅可以提高识别速率，而且还能够满足不同识别速率需求的应用场景。

Description

行人识别装置及方法、辅助驾驶装置 技术领域
本发明属于行人识别技术领域，具体涉及一种行人识别装置及方法以及一种辅助驾驶装置。
背景技术
行人识别是辅助驾驶装置或者智能监控装置的一项关键技术，主要通过图像处理技术实时识别到目标区域是否存在行人，为实时警报提供技术依据。目前，常用的行人识别装置采用摄像头以及计算机视觉技术的方式。实现该方式的硬件设备简单，成本较低，但是实现该方式的软件算法比较复杂。如何提高识别精度和识别速率往往是该领域的关注重点。
对于辅助驾驶装置而言，若画面中出现多个人，在实际应用系统中能识别到一个人即可。因此，识别速率相对识别精度而言更为重要。然而目前常用的行人识别装置是采用滑动窗口遍历搜索图像，识别速率较低，不能够满足要求。
发明内容
本发明旨在至少解决现有技术中存在的技术问题之一，提出了一种行人识别装置及方法以及一种辅助驾驶装置，根据本公开，不仅可以提高识别速率，而且还能够满足不同识别速率需求的应用场景。
为解决上述问题之一，本公开的一个实施例提供了一种行人识别装置，包括：分类器训练电路、图像采集电路、采样电路、计算电路；
所述分类器训练电路，构造成基于训练样本提取行人特征，并基于该行人特征进行离线训练以获得分类器；
所述图像采集电路，构造成采集待识别图像；
所述采样电路，构造成在所述待识别图像上采样获得子图像；
所述计算电路，构造成基于粒子群优化算法识别在所述子图像中 的行人，其中，每个粒子定义为所述子图像中预设大小的对象，并且在粒子群优化算法中基于所述分类器和每个粒子的行人特征计算该粒子的适应度值，所述适应度值表征粒子属于行人的可能性大小。
优选的是，所述行人特征包括积分通道特征。
优选的是，所述分类器包括boost分类器。
优选的是，所述计算电路包括：
初始化子电路，构造成初始化最大迭代次数、粒子数量、每个粒子的位置、速度更新公式中的参数和位置更新公式中的参数；
行人特征提取子电路，构造成提取每个粒子的行人特征；
适应度值计算子电路，构造成基于所述分类器和每个粒子的行人特征计算该粒子的适应度值；
最优位置确定子电路，构造成比较每个粒子的适应度值和该粒子之前搜索位置的适应度值，确定最大值者为该粒子在搜索过程中的局部最优解，并且在所有粒子的局部最优解中，确定适应度值最大值者为全局最优解；
粒子速度更新子电路，构造成基于全局最优解和局部最优解更新每个粒子的速度；
粒子位置更新子电路，构造成基于更新后的粒子速度更新每个粒子的位置，获得下一代粒子；
结果输出子电路，构造成判断所述适应度值计算子电路计算的下一代每个粒子的适应度值是否达到预设值，其中，若是，则所述行人识别装置确定该粒子属于行人；若否，则所述行人识别装置控制所述最优位置确定子电路工作；以及
迭代次数统计子电路，构造成在获得下一代粒子之后当前迭代次数加1，且在迭代次数达到最大迭代次数时向所述行人特征提取子电路发送停止工作指令。
优选的是，所述采样电路，构造成对所述待识别图像按照预设缩放因子进行逐级缩放以获得多级子图像；并且
所述计算电路，构造成在迭代次数达到最大迭代次数时对下一级子图像基于粒子群优化算法识别行人。
本公开的一个实施例还提供一种行人识别方法，包括以下步骤：
基于训练样本提取行人特征，并基于该行人特征离线训练获得分类器；
采集待识别图像；
在所述待识别图像上采样获得子图像；
对所述子图像基于粒子群优化算法识别行人，其中，每个粒子定义为所述子图像中预设大小的对象，并且在所述粒子群优化算法中基于所述分类器和每个粒子的行人特征计算该粒子的适应度值，所述适应度值表示该粒子属于行人的可能性大小。
优选的是，所述行人特征包括积分通道特征。
优选的是，所述分类器包括boost分类器。
优选的是，对所述子图像基于粒子群优化算法识别行人的步骤，包括：
初始化步骤，包括初始化最大迭代次数、粒子数量、每个粒子的位置、速度更新公式中的参数和位置更新公式中的参数；以及
工作步骤：提取每个粒子的行人特征；
基于所述分类器和每个粒子的行人特征计算该粒子的适应度值；
比较每个粒子的适应度值和该粒子之前搜索位置的适应度值，确定最大值者为该粒子在搜索过程中的局部最优解，并且在所有粒子的局部最优解中，确定适应度值最大值者为全局最优解；
基于全局最优解和局部最优解更新每个粒子的速度；
基于更新后的粒子速度更新每个粒子的位置，获得下一代粒子；
提取下一代每个粒子的行人特征，基于所述分类器和下一代的每个粒子的行人特征计算该粒子的适应度值，并判断该适应度值是否达到预设值，其中，若是，则确定该粒子属于行人；若否，则返回所述工作步骤；以及
在获得下一代粒子之后当前迭代次数加1，且在迭代次数达到最大迭代次数时停止所述工作步骤。
优选的是，在所述待识别图像上采样获得子图像的步骤，包括：
对所述待识别图像按照预设缩放因子进行逐级缩放以获得多个子 图像；并且
所述行人识别方法在迭代次数达到最大迭代次数时，还包括：对下一级子图像基于粒子群优化算法识别行人。
本公开的一个实施例还提供一种辅助驾驶装置，包括本发明上述提供的行人识别装置。
另外，本发明的一个实施例还提供了一种电子设备，包括：壳体、处理器、存储器、电路板和电源电路，其中，电路板安置在壳体围成的空间内部，处理器和存储器设置在电路板上；电源电路，用于为上述电子设备的各个电路或器件供电；存储器用于存储可执行程序代码，并且
其中，处理器通过读取存储器中存储的可执行程序代码并运行与可执行程序代码对应的程序，用于执行以下步骤：基于训练样本提取行人特征，并基于该行人特征离线训练获得分类器；
采集待识别图像；
在所述待识别图像上采样获得子图像；
对所述子图像基于粒子群优化算法识别行人，其中，每个粒子定义为所述子图像中预设大小的对象，并且在所述粒子群优化算法中基于所述分类器和每个粒子的行人特征计算该粒子的适应度值，所述适应度值表示该粒子属于行人的可能性大小。
附图说明
图1为根据本发明实施例提供的行人识别装置的原理框图；
图2为图1中计算电路的原理框图；
图3为根据本发明实施例提供的行人识别方法的流程图；
图4为图3中步骤S40的流程图。
具体实施方式
为使本领域的技术人员更好地理解本发明的技术方案，下面结合 附图来对本发明提供的行人识别装置及方法以及一种辅助驾驶装置进行详细描述。
图1为根据本发明实施例提供的行人识别装置的原理框图。请参阅图1，根据本发明实施例提供的行人识别装置包括：分类器训练电路10、图像采集电路20、采样电路30和计算电路40。
其中，分类器训练电路10构造成基于训练样本提取行人特征，并基于该行人特征离线进行训练以获得分类器。图像采集电路20构造成采集待识别图像。采样电路30构造成在待识别图像上采样以获得子图像。计算电路40构造成基于粒子群优化算法识别在子图像中的行人，其中，每个粒子定义为子图像中预设大小的对象；在粒子群优化算法中基于分类器和每个粒子的行人特征计算该粒子的适应度值，适应度值表征粒子属于行人的可能性大小。具体地说，若适应度值越大，则表示属于行人的可能性也就也大；若适应度值越小，则表示属于行人的可能性也就越小。
这里，图像采集电路20可以形成为具有成像功能的摄像头等，计算电路40可以形成为处理器等。
优选的是，行人特征包括但不限于积分通道特征。由于积分通道特征能够更好地描述行人，因此，能够提高识别行人的准确度和识别效率。
具体地说，根据积分通道特征的定义，根据本发明实施例提供的行人识别装置计算出包含LUV(色度和色差)、梯度幅值、方向梯度直方图(Histogram of Gradient简称HOG)特征，并将其顺次连接构成积分通道特征，其中，LUV分别表示亮度和两个色差特征；梯度幅值是上下左右相邻像素差的平方根，梯度方向是上下左右相邻像素差的反正切值(公式(1))，mag表示梯度幅值，Ori表示梯度方向；HOG是每一个像素的梯度幅值在对应的梯度方向上的加权直方图表示(公式(2))：
Figure PCTCN2017107434-appb-000001
Figure PCTCN2017107434-appb-000002
其中，公式(1)和(2)中的fi，j表示图像位置(i，j)处的像素值；公式(2)中的
Figure PCTCN2017107434-appb-000003
表示第
Figure PCTCN2017107434-appb-000004
个直方图。
优选的是，分类器包括但不限于boost分类器。boost分类器是将若干个弱分类器强化成一个强分类器的过程。因此，boost分类器具有使用灵活，分类精度高的特点，同时还兼顾了计算性能。
具体地说，请参阅图2，计算电路40包括：初始化子电路401、行人特征提取子电路402、适应度计算子电路403、最优位置确定子电路404、粒子速度更新子电路405、粒子位置更新子电路406、结果输出子电路407和迭代次数统计子电路408。
其中，初始化模块401构造成初始化最大迭代次数、粒子数量、每个粒子的位置、速度更新公式中的参数和位置更新公式中的参数。
在此需要说明的是，最大迭代次数和粒子数量通常为人为自行设定的；初始化的每个粒子的位置通常设置为随机值；速度更新公式和位置更新公式中的参数可以为根据经验获得的最优常值。对于行人识别而言，搜索空间为二维空间。因此，粒子的位置可以用(xi1，xi2)表示。每个粒子不仅具有位置属性，还具有速度属性和适应度值属性。
行人特征提取子电路402构造成提取每个粒子的行人特征。
适应度值计算子电路403构造成基于分类器和每个粒子的行人特征计算该粒子的适应度值。具体地说，适应度值用于衡量粒子优劣，同时也作为确定局部和全局最优粒子的依据，更是粒子搜索的基础依据。
最优位置确定子电路404构造成比较每个粒子的适应度值和该粒子之前搜索位置的适应度值，确定最大值者为该粒子在搜索过程中的局部最优解，并且在所有的粒子的局部最优解中，确定适应度值最大值者为全局最优解。
粒子速度更新子电路405构造成基于全局最优解和局部最优解更新每个粒子的速度。具体地说，由于所有粒子的运动会受到局部最优解和全局最优解的影响，因此，速度更新公式包括但不限于以下方式：
Figure PCTCN2017107434-appb-000005
其中，ω是保持原来速度的系数，所以叫做惯性权重，通常设置为0.79；
c1是粒子跟踪自己历史最优值的权重系数，它表示粒子自身的认知，所以叫认知系数，通常设置为2；
c2是粒子跟踪群体最优值的权重系数，它表示粒子对整个群体知识的认知，所以叫做社会系数，通常设置为2；
ξ、η是[0，1]区间内均匀分布的随机数，其作用是产生随机扰动，通常设置为0.7；
d表示维度，对于行人识别只有x和y两个维度；
Figure PCTCN2017107434-appb-000006
表示第i个粒子的d维度在k次迭代的局部最优解；
Figure PCTCN2017107434-appb-000007
表示所有粒子的d维度在k次迭代的全局最优解；
Figure PCTCN2017107434-appb-000008
表示第i个粒子的d维度在k次迭代的位置。
速度更新公式的物理意义为：第i个粒子的d维度在k+1次迭代中的速度＝该粒子在k次迭代中的速度与惯性系数的积+该粒子d维度受局部最优解的吸引量+该粒子d维度受全局最优解的吸引量。通常，为防止计算过大，会设置速度的最大值与最小值。
粒子位置更新子电路406构造成基于更新后的粒子速度更新每个粒子的位置，获得下一代粒子。具体地说，因为每个粒子具有上文中计算得到的速度，因此粒子的位置会在该速度的驱动下按照如下公式更新：
Figure PCTCN2017107434-appb-000009
其中，r为速度约束量，通常设置为1。
位置更新公式的物理意义为：第i个粒子在k+1次迭代搜索中在d维度的位置＝上次在d维度的位置+速度与约束量的积。
需要在此说明的是，在粒子更新过程中，需要粒子始终保持在搜索空间范围内。对于[m，n]大小的子图像，粒子大小为64*128，因此搜索空间为：[m-64，n-128]。
结果输出子电路407构造成判断适应度值计算子电路403计算的下一代每个粒子的适应度值是否达到预设值，其中，若是，则所述行 人识别装置确定该粒子属于行人；若否，则所述行人识别装置控制所述最优位置确定子电路404继续工作。可以理解，通过使用预设值作为阈值来判断粒子是否属于行人，可以进一步提高识别速率。
在实际应用中，具体地说，若结果输出子电路确定该粒子属于行人，则在输出图像中叠加该粒子对应的图像，和/或，提供报警。
迭代次数统计子电路408构造成在获得下一代粒子之后当前迭代次数累积加1，且在迭代次数达到最大迭代次数时，向行人特征提取子电路402发送停止工作指令。
进一步具体地说，采样电路30构造成对待识别图像按照预设缩放因子进行逐级缩放以获得多级子图像；并且计算电路40构造成在迭代次数达到最大迭代次数时对下一级子图像基于粒子群优化算法识别行人。通过采样电路30对待识别图像按照预设缩放因子进行逐级缩放以获得多级子图像，这样，可以在子图像中行人较小时在放大的子图像中识别出行人，在子图像中行人较大时在缩小的子图像中识别出行人。因此，根据本发明实施例的行人识别装置可以快速地识别出行人。
更具体地说，假设一幅图像大小为[m，n]，采样公式可以表示为：scale*[m，n]，其中scale为缩放因子。如果缩放因子大于1，采样的每级子图像宽度和高度会相对上一级子图像按照比例放大。此时，子图像可称之为进行向上采样。如果0＜scale＜1，采样的每级子图像宽度和高度会相对上一级子图像按照比例缩小。此时，子图像可称之为进行向下采样。这里，得到的多级子图像可依次称之为第1级子图像、第2级子图像……，其中，第1级子图像为第2级子图像的上一级子图像。
下面详细描述本发明实施例提供的行人识别装置的工作过程。具体地说，所述工作过程包括以下步骤：
S1，初始化子电路401初始化最大迭代次数、粒子数量、每个粒子的位置、速度更新公式中的参数和位置更新公式中的参数；
S2，在视频中选取一帧图像作为待识别图像；以及
S3，采样电路30对待识别图像按照预设缩放因子进行逐级缩放以获得多级子图像。
在选取第1级子图像时，所述工作过程还执行以下步骤：
S4，行人特征提取子电路402提取每个粒子的行人特征；
S5，适应度值计算子电路403基于分类器和每个粒子的行人特征计算该粒子的适应度值；
S6，最优位置确定子电路404确定出局部最优解和全局最优解；
S7，粒子速度更新子电路405基于局部最优解和全局最优解更新每个粒子的速度；
S8，粒子位置更新子电路406基于更新后的粒子速度更新每个粒子的位置，获得下一代粒子。
S9，行人特征提取子电路402提取下一代粒子的行人特征；
S10，适应度值计算子电路403基于分类器和下一代每个粒子的行人特征计算该粒子的适应度值；
S11，结果输出子电路407判断下一代每个粒子的适应度值是否达到预设值，其中，若是，则所述行人识别装置确定该粒子属于行人，在输出图像中叠加该粒子对应的图像，和/或，提供报警；若否，则所述行人识别装置执行步骤S12然后返回步骤S6；以及
S12，迭代次数统计子电路408在获得下一代粒子之后，将当前迭代次数加1，并且在迭代次数达到最大迭代次数时对下一级子图像执行步骤S4。
若针对多级子图像在步骤S10和S11中均没有识别到行人时，进入步骤S2，将视频的下一帧图像作为待识别图像，然后继续工作。
综上所述，本发明实施例提供的行人识别装置，将粒子群优化算法引用了行人识别领域，将一个粒子广义化为以该粒子为起点大小为预设大小的行人识别窗口，以及将适应度值广义化为表征属于行人的概率得分。通过粒子群优化(Particle Swarm Optimization，PSO)算法在搜索图像中能快速收敛到概率得分较高的区域，并且，可以使得行人识别的识别速率不再与图像大小成正比例，而是与优化算法的粒子数相关联，从而能够满足不同识别速率需求的应用场景。
图3为本发明实施例提供的行人识别方法的流程图。请参阅图3，本发明实施例提供的行人识别方法包括以下步骤：
S10，基于训练样本提取行人特征，并基于该行人特征离线训练获得分类器；
S20，采集待识别图像；
S30，在所述待识别图像上采样获得子图像；
S40，对所述子图像基于粒子群优化算法识别行人，其中，每个粒子定义为子图像中预设大小的对象，并且在所述粒子群优化算法中基于所述分类器和每个粒子的行人特征计算该粒子的适应度值，适应度值表示该粒子属于行人的可能性大小。
优选的是，行人特征包括积分通道特征。
优选的是，分类器包括boost分类器。
请参阅图4，上述步骤S40包括以下步骤：
S401，初始化步骤，包括初始化最大迭代次数、粒子数量、每个粒子的位置、速度更新公式中的参数和位置更新公式中的参数；以及
S402，工作步骤，包括：
S4021，提取每个粒子的行人特征；
S4022，基于分类器和每个粒子的行人特征计算该粒子的适应度值；
S4023，比较每个粒子的适应度值和该粒子之前搜索位置的适应度值，确定最大值者为该粒子在搜索过程中的局部最优解，并且在所有粒子的局部最优解中，确定适应度值最大值者为全局最优解；
S4024，基于全局最优解和局部最优解更新每个粒子的速度；
S4025，基于更新后的粒子速度更新每个粒子的位置，获得下一次粒子；
S4026，提取下一代每个粒子的行人特征，基于所述分类器和下一代每个粒子的行人特征计算该粒子的适应度值，并判断该适应度值是否达到预设值，其中，若是，则确定该粒子属于行人；若否，则返回该工作步骤；以及
S4027，在获得下一代粒子之后当前迭代次数加1，且在迭代次数达到最大迭代次数时该停止工作步骤。
更具体地说，步骤S30包括：对待识别图像按照预设缩放因子进 行逐级缩放以获得多个子图像。在这种情况下，该行人识别方法在迭代次数达到最大迭代次数时，步骤S4022还包括：对下一级子图像基于粒子群优化算法识别行人。
由于本发明实施例提供的行人识别方法与本发明上述实施例提供的行人识别装置相对应，而行人识别装置在上文中已经有了详细地描述，因此，本实施例提供的行人识别方法的相关内容在此不再详述，请见行人识别装置中的相应内容。
综上所述，根据本发明实施例提供的行人识别方法，将粒子群优化算法引用了行人识别领域，将一个粒子广义化为以该粒子为起点大小为预设大小的行人识别窗口，以及将适应度值广义化为表征属于行人的概率得分。通过粒子群优化(PSO)算法在搜索图像中能快速收敛到概率得分较高的区域，并且，可以使得行人识别的识别速率不再与图像大小成正比例，而是与优化算法的粒子数相关联，从而能够满足不同识别速率需求的应用场景。
本发明的一个实施例还提供一种辅助驾驶装置，包括上述第一实施例提供的行人识别装置。所述辅助驾驶装置可以包括摄像头，处理器，存储器等。所述处理器可以与导航系统或扩视系统一起集成于中控板、后视镜或行车记录设备等结构中。这里，所述辅助驾驶装置还包括辅助司机驾驶装置。此外，所述辅助驾驶装置还包括无人驾驶汽车等中的辅助装置。
本发明实施例提供的辅助驾驶装置，由于采用本发明上述第一实施例提供的行人识别装置，因此能够快速地识别到行人。因此，能够获得良好的辅助驾驶效果，并且应用性强。
本发明的一个实施例还提供一种电子设备，包括：壳体、处理器、存储器、电路板和电源电路，其中，电路板安置在壳体围成的空间内部，处理器和存储器设置在电路板上；电源电路，用于为上述电子设备的各个电路或器件供电；存储器用于存储可执行程序代码。处理器通过读取存储器中存储的可执行程序代码并运行与可执行程序代码对应的程序，用于执行以下步骤：基于训练样本提取行人特征，并基于 该行人特征离线训练获得分类器；
采集待识别图像；
在所述待识别图像上采样获得子图像；
对所述子图像基于粒子群优化算法识别行人，其中，每个粒子定义为所述子图像中预设大小的对象，并且在所述粒子群优化算法中基于所述分类器和每个粒子的行人特征计算该粒子的适应度值，所述适应度值表示该粒子属于行人的可能性大小。
这里，该电子设备可以集成在车辆的中控板、后视镜或行车记录设备等结构中，或者独立于车辆的中控板、后视镜或行车记录设备。
可以理解的是，以上实施方式仅仅是为了说明本发明的原理而采用的示例性实施方式，然而本发明并不局限于此。对于本领域内的普通技术人员而言，在不脱离本发明的精神和实质的情况下，可以做出各种变型和改进，这些变型和改进也视为本发明的保护范围。

Claims (14)

  1. 一种行人识别装置，其中，所述行人识别装置包括：分类器训练电路和计算电路，其中
    所述分类器训练电路，构造成基于训练样本提取行人特征，并基于该行人特征进行离线训练以获得分类器；并且
    所述计算电路，构造成基于粒子群优化算法识别在图像中的行人，其中，每个粒子定义为所述图像中预设大小的对象，并且在粒子群优化算法中基于所述分类器和每个粒子的行人特征计算该粒子的适应度值，所述适应度值表征粒子属于行人的可能性大小。
  2. 根据权利要求1所述的行人识别装置，还包括：图像采集电路和采样电路，其中
    所述图像采集电路构造成采集待识别图像；并且
    所述采样电路，构造在所述待识别图像上采样获得子图像。
  3. 根据权利要求2所述的行人识别装置，其中
    所述计算电路构造成基于粒子群优化算法识别在所述子图像中的行人，其中，每个粒子定义为所述子图像中预设大小的对象，并且在粒子群优化算法中基于所述分类器和每个粒子的行人特征计算该粒子的适应度值，所述适应度值表征粒子属于行人的可能性大小。
  4. 根据权利要求1所述的行人识别装置，其中，所述行人特征包括积分通道特征。
  5. 根据权利要求1所述的行人识别装置，其中，所述分类器包括boost分类器。
  6. 根据权利要求1或2所述的行人识别装置，其中，所述计算电路包括：
    初始化子电路，构造成初始化最大迭代次数、粒子数量、每个粒子的位置、速度更新公式中的参数和位置更新公式中的参数；
    行人特征提取子电路，构造成提取每个粒子的行人特征；
    适应度值计算子电路，构造成基于所述分类器和每个粒子的行人特征计算该粒子的适应度值；
    最优位置确定子电路，构造成比较每个粒子的适应度值和该粒子之前搜索位置的适应度值，确定最大值者为该粒子在搜索过程中的局部最优解，并且在所有粒子的局部最优解中，确定适应度值最大值者为全局最优解；
    粒子速度更新子电路，构造成基于全局最优解和局部最优解更新每个粒子的速度；
    粒子位置更新子电路，构造成基于更新后的粒子速度更新每个粒子的位置，获得下一代粒子；
    结果输出子电路，构造成判断所述适应度值计算子电路计算的下一代每个粒子的适应度值是否达到预设值，其中，若是，则所述行人识别装置确定该粒子属于行人；若否，则所述行人识别装置控制所述最优位置确定子电路工作；以及
    迭代次数统计子电路，构造成在获得下一代粒子之后当前迭代次数加1，且在迭代次数达到最大迭代次数时，向所述行人特征提取子电路发送停止工作指令。
  7. 根据权利要求6所述的行人识别装置，其中，
    所述采样电路，构造成对所述待识别图像按照预设缩放因子进行逐级缩放以获得多级子图像；并且
    所述计算电路，构造成在迭代次数达到最大迭代次数时对下一级子图像基于粒子群优化算法识别行人。
  8. 一种行人识别方法，其中，包括以下步骤：
    基于训练样本提取行人特征，并基于该行人特征离线训练获得分类器；
    采集待识别图像；
    在所述待识别图像上采样获得子图像；
    对所述子图像基于粒子群优化算法识别行人，其中，每个粒子定义为所述子图像中预设大小的对象，并且在所述粒子群优化算法中基于所述分类器和每个粒子的行人特征计算该粒子的适应度值，所述适应度值表示该粒子属于行人的可能性大小。
  9. 根据权利要求8所述的行人识别方法，其中，所述行人特征包括积分通道特征。
  10. 根据权利要求8所述的行人识别方法，其中，所述分类器包括boost分类器。
  11. 根据权利要求8所述的行人识别方法，其中，对所述子图像基于粒子群优化算法识别行人的步骤，包括：
    初始化步骤，包括初始化最大迭代次数、粒子数量、每个粒子的位置、速度更新公式中的参数和位置更新公式中的参数；以及
    工作步骤，包括：
    提取每个粒子的行人特征；
    基于所述分类器和每个粒子的行人特征计算该粒子的适应度值；
    比较每个粒子的适应度值和该粒子之前搜索位置的适应度值，确定最大值者为该粒子在搜索过程中的局部最优解，并且在所有粒子的局部最优解中，确定适应度值最大值者为全局最优解；
    基于全局最优解和局部最优解更新每个粒子的速度；
    基于更新后的粒子速度更新每个粒子的位置，获得下一代粒子；
    提取下一代每个粒子的行人特征，基于所述分类器和下一代的每个粒子的行人特征计算该粒子的适应度值，并判断该适应度值是否达到预设值，其中，若是，则确定该粒子属于行人；若否，则返回所述工作步骤；以及
    在获得下一代粒子之后当前迭代次数加1，且在迭代次数达到最大 迭代次数时停止所述工作步骤。
  12. 根据权利要求11所述的行人识别方法，其中，在所述待识别图像上采样获得子图像的步骤，包括：
    对所述待识别图像按照预设缩放因子进行逐级缩放以获得多个子图像；并且
    所述行人识别方法在迭代次数达到最大迭代次数时，还包括：对下一级子图像基于粒子群优化算法识别行人。
  13. 一种辅助驾驶装置，其中，所述辅助驾驶装置包括权利要求1-7中任意一项所述的行人识别装置。
  14. 一种电子设备，包括：壳体、处理器、存储器、电路板和电源电路，其中，电路板安置在壳体围成的空间内部，处理器和存储器设置在电路板上；电源电路，用于为上述电子设备的各个电路或器件供电；存储器用于存储可执行程序代码，并且
    其中，处理器通过读取存储器中存储的可执行程序代码并运行与可执行程序代码对应的程序，用于执行以下步骤：基于训练样本提取行人特征，并基于该行人特征离线训练获得分类器；
    采集待识别图像；
    在所述待识别图像上采样获得子图像；
    对所述子图像基于粒子群优化算法识别行人，其中，每个粒子定义为所述子图像中预设大小的对象，并且在所述粒子群优化算法中基于所述分类器和每个粒子的行人特征计算该粒子的适应度值，所述适应度值表示该粒子属于行人的可能性大小。
PCT/CN2017/107434 2017-04-10 2017-10-24 行人识别装置及方法、辅助驾驶装置 WO2018188309A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP17882265.6A EP3611652A4 (en) 2017-04-10 2017-10-24 PEDESTRIAN IDENTIFICATION DEVICE AND METHOD, AND DRIVING ASSISTANCE DEVICE
US16/066,799 US11508157B2 (en) 2017-04-10 2017-10-24 Device and method of objective identification and driving assistance device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710229280.7A CN108694356B (zh) 2017-04-10 2017-04-10 行人检测装置及方法、辅助驾驶系统
CN201710229280.7 2017-04-10

Publications (1)

Publication Number Publication Date
WO2018188309A1 true WO2018188309A1 (zh) 2018-10-18

Family

ID=63793107

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/107434 WO2018188309A1 (zh) 2017-04-10 2017-10-24 行人识别装置及方法、辅助驾驶装置

Country Status (4)

Country Link
US (1) US11508157B2 (zh)
EP (1) EP3611652A4 (zh)
CN (1) CN108694356B (zh)
WO (1) WO2018188309A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109711366A (zh) * 2018-12-29 2019-05-03 浙江大学 一种基于群组信息损失函数的行人重识别方法
CN114459501A (zh) * 2021-12-10 2022-05-10 华中光电技术研究所（中国船舶重工集团公司第七一七研究所） 一种惯导安装误差自动标定方法及装置
CN116503399A (zh) * 2023-06-26 2023-07-28 广东电网有限责任公司湛江供电局 基于yolo-afps的绝缘子污闪检测方法

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110444022A (zh) * 2019-08-15 2019-11-12 平安科技（深圳）有限公司 交通流数据分析模型的构建方法和装置
CN111461177B (zh) * 2020-03-09 2023-07-07 北京邮电大学 一种图像的识别方法及装置
CN114022538A (zh) * 2021-09-15 2022-02-08 苏州中科华影健康科技有限公司 一种内窥镜的路径规划方法、装置、终端及存储介质
CN115100365B (zh) * 2022-08-25 2023-01-20 国网天津市电力公司高压分公司 一种基于粒子群算法的相机最优基线获取方法
CN117152258B (zh) * 2023-11-01 2024-01-30 中国电建集团山东电力管道工程有限公司 一种管道生产智慧车间的产品定位方法及系统

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8768868B1 (en) * 2012-04-05 2014-07-01 Hrl Laboratories, Llc Optimal multi-class classifier threshold-offset estimation with particle swarm optimization for visual object recognition
CN103927531A (zh) * 2014-05-13 2014-07-16 江苏科技大学 一种基于局部二值和粒子群优化bp神经网络的人脸识别方法
CN106485283A (zh) * 2016-10-19 2017-03-08 电子科技大学 一种基于Online Boosting的粒子滤波行人目标跟踪方法

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8589315B2 (en) * 2004-08-14 2013-11-19 Hrl Laboratories, Llc Behavior recognition using cognitive swarms and fuzzy graphs
US8301406B2 (en) * 2008-07-24 2012-10-30 University Of Cincinnati Methods for prognosing mechanical systems
WO2010088465A1 (en) * 2009-02-02 2010-08-05 Gentex Corporation Improved digital image processing and systems incorporating the same
US10417486B2 (en) * 2013-12-30 2019-09-17 Alcatel Lucent Driver behavior monitoring systems and methods for driver behavior monitoring
US11436395B2 (en) * 2018-06-27 2022-09-06 Dalian University Of Technology Method for prediction of key performance parameter of an aero-engine transition state acceleration process based on space reconstruction
CN110333325B (zh) * 2019-08-02 2021-09-17 中南大学 一种大气污染环境下列车运行防护方法及系统

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8768868B1 (en) * 2012-04-05 2014-07-01 Hrl Laboratories, Llc Optimal multi-class classifier threshold-offset estimation with particle swarm optimization for visual object recognition
CN103927531A (zh) * 2014-05-13 2014-07-16 江苏科技大学 一种基于局部二值和粒子群优化bp神经网络的人脸识别方法
CN106485283A (zh) * 2016-10-19 2017-03-08 电子科技大学 一种基于Online Boosting的粒子滤波行人目标跟踪方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
See also references of EP3611652A4 *
WANG JUN: "Design and implementation of vehicle infrared pedestrian detection system", MASTER THESIS, 15 March 2017 (2017-03-15), pages 1 - 71, XP009515758, ISSN: 1674-0246 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109711366A (zh) * 2018-12-29 2019-05-03 浙江大学 一种基于群组信息损失函数的行人重识别方法
CN114459501A (zh) * 2021-12-10 2022-05-10 华中光电技术研究所（中国船舶重工集团公司第七一七研究所） 一种惯导安装误差自动标定方法及装置
CN114459501B (zh) * 2021-12-10 2023-08-25 华中光电技术研究所（中国船舶重工集团公司第七一七研究所） 一种惯导安装误差自动标定方法及装置
CN116503399A (zh) * 2023-06-26 2023-07-28 广东电网有限责任公司湛江供电局 基于yolo-afps的绝缘子污闪检测方法
CN116503399B (zh) * 2023-06-26 2024-03-22 广东电网有限责任公司湛江供电局 基于yolo-afps的绝缘子污闪检测方法

Also Published As

Publication number Publication date
EP3611652A4 (en) 2021-01-06
EP3611652A1 (en) 2020-02-19
CN108694356B (zh) 2024-05-07
CN108694356A (zh) 2018-10-23
US20210166042A1 (en) 2021-06-03
US11508157B2 (en) 2022-11-22

Similar Documents

Publication Publication Date Title
WO2018188309A1 (zh) 行人识别装置及方法、辅助驾驶装置
US11393103B2 (en) Target tracking method, device, system and non-transitory computer readable medium
CN107529650B (zh) 闭环检测方法、装置及计算机设备
WO2018219198A1 (zh) 一种人机交互方法、装置及人机交互终端
CN108960189B (zh) 图像再识别方法、装置及电子设备
JP4677046B2 (ja) 多次元ブースト回帰を経た外観及び動作を使用する高速人間姿勢推定
JP2021120864A (ja) 障害物を検出するための方法および装置、電子機器、記憶媒体並びにコンピュータプログラム
JP6482195B2 (ja) 画像認識装置、画像認識方法及びプログラム
JP2016062610A (ja) 特徴モデル生成方法及び特徴モデル生成装置
WO2012139273A1 (en) Method of detecting facial attributes
JP2009026314A (ja) 複数の外見モデルを用いた多姿勢顔追跡
JP6756406B2 (ja) 画像処理装置、画像処理方法および画像処理プログラム
CN103443804A (zh) 面部标志检测方法
WO2022121130A1 (zh) 电力目标检测方法、装置、计算机设备和存储介质
WO2021084972A1 (ja) 物体追跡装置および物体追跡方法
CN109800682A (zh) 驾驶员属性识别方法及相关产品
CN113255557B (zh) 一种基于深度学习的视频人群情绪分析方法及系统
JP2021503139A (ja) 画像処理装置、画像処理方法および画像処理プログラム
CN112561879B (zh) 模糊度评价模型训练方法、图像模糊度评价方法及装置
CN113743426A (zh) 一种训练方法、装置、设备以及计算机可读存储介质
CN108875488B (zh) 对象跟踪方法、对象跟踪装置以及计算机可读存储介质
CN112700568A (zh) 一种身份认证的方法、设备及计算机可读存储介质
CN116434156A (zh) 目标检测方法、存储介质、路侧设备及自动驾驶系统
Chang et al. Robust abandoned object detection and analysis based on online learning
CN112749727A (zh) 本地服务器、影像辨识系统及其更新方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17882265

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2017882265

Country of ref document: EP

Effective date: 20191111