A kind of intelligent road-lamp of built-in collision sound detection functionTechnical field
The present invention relates to intelligent transportation field, more particularly to a kind of built-in collision sound detection function of being applied to road lightingIntelligent road-lamp.
Background technology
In recent years, with cloud computing, big data, artificial intelligence, the especially maturation of high-performance calculation hardware technology, machineThe maturation of learning areas deep learning (deep learning, DL) theory and automated characterization learning neural network model, depth, in artificial intelligence field (intelligent image and voice recognition) extensive use, major IT giants are by artificial intelligence for neutral netAs next core technology growth point, meanwhile, the integrated core of large quantities of build-in depths neutral net frameworks has also been expedited the emergence of in marketPiece, such as the tall and handsome chip Tesla P100 that first design of starting from scratch exclusively for deep learning is released up to (NVIDIA), the coreSheet data processing speed is release GPU series for its 2014 12 times；Google is the chip TPU of machine learning customization by hardwareCan be promoted to equivalent to the level after developing 7 years according to Moore's Law；In artificial intelligence field, also unwillingness falls domestic relevant enterpriseAfterwards, such as Chengdu is opened Ying Tailun science and technology and releases a intelligent sound chip CI1006, is the artificial intelligence language based on ASIC frameworksSound identification chip, contains deep neural network treatment hardware cell, perfect can support DNN computing frameworks, carries out high-performanceData parallel, can greatly improve treatment effeciency of the artificial intelligence deep learning voice technology to mass data；CI1006 can reduce dependence of the product for network, lifting intelligent sound identification response using local Neural Network Data treatmentAnd control speed, can be widely applied to the product scopes such as intelligent household appliances, robot, wisdom toy.
Therefore, by deep neural network be implanted in existing embedded intelligence control system technically completely intoIt is ripe.Although the street lamp of prior art has formed intelligent control network, its intellectuality for only realizing road lighting, and street lampSystem, if integrated security monitoring function, will be able to play street lamp bigger as a huge network in physical spaceEffect, such as major traffic accidents monitoring.When major traffic accidents occur, travel condition of vehicle have sent corresponding change,Especially when high velocity impact, with the sound of sharp impacts, by recognizing that impact sound can just detect traffic accident.SoAnd, ambient noise is noisy in road, and impact sound cannot be accurately detected using existing sound detection technology.
Therefore, for drawbacks described above present in currently available technology, it is necessary to be studied in fact, to provide a kind of scheme,Solve defect present in prior art.
The content of the invention
In view of this, it is necessory to provide the built-in intelligent road-lamp for colliding sound detection function such that it is able to quick detection weightBig traffic accident, timely early warning, rescue, it is to avoid accident is sent out after causing.
In order to overcome the defect of prior art, technical scheme is as follows：
A kind of intelligent road-lamp of built-in collision sound detection function, sets sound acquisition module, impact sound identification in street lampModule, memory module and for Street lamps control and the controller for road lamp of information transfer；
The voice signal that the sound acquisition module is used in continuous acquisition road；
The memory module is used to store the voice signal that the sound acquisition module is gathered；
Whether the impact sound identification module is used to recognizing in gathered voice signal miscellaneous containing impact sound and will identification knotFruit is sent to controller for road lamp, and whether the controller for road lamp judges anomalous event occur in road and will send information to thisRemote monitoring center；
The impact sound identification module includes the extraction module of feature first, the first normalization module, neural network classification mouldBlock and neural metwork training module, wherein,
The fisrt feature extraction module is used to receive acquired original voice data, and acquired original voice data is carried outFeature extraction；
The first normalization module is used to carry out the data after feature extraction Gaussian normalization processing, output normalizationData；
The neural network classification module is used to receive normalization data and the good deep neural network of training in advance, and leads toCrossing deep neural network carries out Classification and Identification acquisition sorting result information to the normalization data, and the sorting result information isWhether the miscellaneous probable value containing impact sound of acquired original voice data, when the probable value exceedes threshold value set in advance, then judgesOccurs anomalous event in road；
The neural metwork training module is used to receive training data and the training of neutral net is carried out according to training data,Obtain abundant learning tape to make an uproar the deep neural network of the Nonlinear Mapping relation between sample and clean sample, and export the depthSpend neutral net to the neural network classification module；
The neural metwork training module includes that second feature extraction module, the second normalization module, unsupervised learning are pre-Training module and supervised learning optimization module, wherein, the unsupervised learning pre-training module is used to find depth in input dataLayer abstract characteristics, using be restricted Boltzmann machine (RBM) model carry out pre-training and by unsupervised learning by the way of byLayer progressive learning neural network parameter；The supervised learning optimization module uses backpropagation (back-propagation, BP)Algorithm, the intense adjustment for having supervision is carried out to neural network parameter using labeled data.
Preferably, the fisrt feature extraction module/second feature extraction module further includes that framing module, DFT becomeMold changing block and log power spectrum processing module, wherein, the framing module is used to carry out sub-frame processing to input data；It is describedDFT transform module obtains frequency domain information for carrying out discrete Fourier transform to the data after sub-frame processing；The log powerSpectrum processing module is used to carry out log power spectrum treatment to frequency domain information.
Preferably, the impact sound identification module also includes average energy detection module and frame energy comparison module, wherein,The average energy detection module is used to calculate the average energy value of present frame log power spectrum and be sent to the frame energy ratioCompared with module；Whether the frame energy comparison module is used to calculate the difference of consecutive frame the average energy value and judges the difference beyond pre-If threshold value, if it exceeds then opening the neural network classification module.
Preferably, the frame energy comparison module receives current information of vehicle flowrate, and calculates present frame the average energy value and beThe no energy value scope beyond corresponding to default corresponding information of vehicle flowrate, if it exceeds then opening the neural network classification mouldBlock.
Preferably, at the sound acquisition module further includes to be made up of multiple microphones microphone array, audioReason module and control module, wherein,
It is multiple in the microphone array that microphone is in certain geometrical shape and each microphone has unique ID；
The audio processing modules are used to synchronously obtain and identify the voice signal of each microphone collection and to the soundMessage number exports audio-frequency information after being processed；
The control module is connected with the audio processing modules, for control the audio processing modules work simultaneouslyAudio-frequency information after audio processing modules treatment is stored in the memory module.
Preferably, also including impact sound locating module, the impact sound locating module is connected with the controller for road lamp,For obtaining the particular location of impact sound and positional information being sent into the controller for road lamp；
When the impact sound identification module judges anomalous event occur, the impact sound locating module obtains the abnormal thingThe temporal information of part simultaneously obtains the audio-frequency information that corresponding microphone array is listed in the temporal information from the memory module, and rootThe parameter information of the position relationship fixed according to each microphone and each microphone correspondence audio-frequency information in the temporal informationDetermine the particular location residing for impact sound.
Preferably, the parameter information be each microphone in the temporal information correspondence audio-frequency information peak strength withAnd each microphone is in the peak strength corresponding time difference.
Preferably, the controller for road lamp is also connected with Rotatable camera device, and the Rotatable camera device is setRotated on light pole and according to the control instruction of the controller for road lamp；
When the impact sound identification module judges anomalous event occur in road, the controller for road lamp control is described canRotating pick-up device is rotated to particular location determined by the impact sound locating module.
Preferably, the multiple microphones in the microphone array are arranged on lamp surface with certain geometrical shape.
Preferably, the impact sound identification module is realized using the artificial intelligence chip of build-in depths neutral net.
Compared with prior art, be integrated in impact sound identification technology in street lamp and road monitoring neck is applied to by the present inventionDomain, realizes that impact sound is recognized using deep neural network, so that impact sound accuracy of identification is improved, the monitoring combined using audio frequency and videoThe further perfect conduct monitoring at all levels of road safety of mode, and can be in the road traffic of early warning in time anomalous event.
Brief description of the drawings
Fig. 1 is the theory diagram of the intelligent road-lamp of the built-in collision sound detection function of the present invention.
Fig. 2 is the theory diagram of impact sound identification module in the built-in intelligent road-lamp for colliding sound detection function of the present invention.
Fig. 3 is to be restricted Boltzmann machine (RBM) structural representation.
Fig. 4 is the pre-training schematic diagram of RBM in the present invention.
Fig. 5 is the structured flowchart of the deep neural network that training is obtained.
Fig. 6 is the theory diagram of characteristic extracting module in the present invention.
Fig. 7 is the theory diagram of impact sound identification module another embodiment in the present invention.
Fig. 8 is the theory diagram of sound acquisition module in the present invention.
Fig. 9 is the schematic diagram of microphone array arrangement.
Figure 10 is the theory diagram of another preferred embodiment of the invention.
Figure 11 is the Organization Chart of the intelligent sound Processing with Neural Network chip CI1006 that the present invention is used.
Specific examples below will further illustrate the present invention with reference to above-mentioned accompanying drawing.
Specific embodiment
The intelligent road-lamp of the built-in collision sound detection function of being provided the present invention below with reference to accompanying drawing is described further.
As background technology is introduced, machine learning field deep learning (deep learning, DL) it is theoretical and fromThe maturation of dynamic feature learning neural network model, deep neural network in artificial intelligence field (intelligent image and voice recognition)Through extensive use, the speech recognition technology of prior art can recognize the voice of people, Er Qieneng under complicated Background environmental noiseRelatively accurately identify semanteme.In technical field of voice recognition, recognized whether under complicated Background environmental noise voice (orPerson other sound) and in the absence of technical difficulty, and real difficulty is semantics recognition, the conversion of voice accuracy high is writtenWord not enough, will understand what the mankind saying, what to be expressed and is intended to, and this is only the jewel on imperial crown.This is primarily due toThe species of voice has almost countless and different people to be even more different accents, while the ambient noise ring residing for voiceBorder is even more changeable, and almost each voice scene can have different ambient noises.Accordingly, it would be desirable to huge amount of calculation could be completedReal-time semantic analysis.
Relative to the applied environment of the application, although ambient noise is complex in road, but car crass sound typeIt is to be relatively fixed, the especially impact sound produced by high velocity impact, the complexity of impact sound identification is not as good as the ten thousand of voice complexity/ mono-, while impact sound instantaneous strength is greatly, the easy subregion of sound characteristic.Using the intelligent sound of deep neural network (DNN)Sound identification is that conventional acoustic treatment technology needs to assume various preferable shapes relative to the advantage of conventional acoustic treatment technologyState, the desirability that these are assumed turns into the key factor of influence performance naturally, and DNN is with little need for any other conditionIt is assumed that can constantly be approached by constantly study, so as to reach the purpose of accurate identification.That is DNN by multilayer and it is thousands of onTen thousand neuron nodes with computing capability are superimposed as a depth network structure, then this DNN are trained, with bigRoad noise training DNN under amount clean sample and various situations, the purpose is to know to enough from known data learningKnow, be then generalized to following emerging data, make effective decision-making.Namely made an uproar sample and clean as learning tape with DNNThe regression model of the Nonlinear Mapping relation between sample, using the depth structure and non-linear simulation ability of DNN, Ke YichongThe complicated interaction relationship for dividing learning tape to make an uproar between sample and clean sample, the learning process of neutral net is one unlimitedThe process of approaching to reality, can automatically adjust the parameter and weight of neutral net according to the data of input, and the data of its training are more,The result of identification is more accurate.After DNN training is completed, when actually detected, the road acoustical signal input that will be gathered in real roadIn DNN, so as to judge whether the voice signal is miscellaneous containing impact sound.
Referring to Fig. 1, the theory diagram of the intelligent road-lamp of the built-in collision sound detection function of the present invention is shown, set in street lampPut sound acquisition module, impact sound identification module, memory module and for Street lamps control and the controller for road lamp of information transfer；The voice signal that sound acquisition module is used in continuous acquisition road；Memory module is used to store what sound acquisition module was gatheredVoice signal；Whether impact sound identification module is used to recognizing in gathered voice signal miscellaneous sends out containing impact sound and by recognition resultController for road lamp is sent to, whether controller for road lamp judges anomalous event occur in road and will send information to remote monitoring with thisCenter.
Referring to Fig. 2, the original of impact sound identification module in the intelligent road-lamp of the built-in collision sound detection function of the present invention is shownReason block diagram, including the extraction module of feature first, the first normalization module, neural network classification module and neural metwork training mouldBlock, wherein, fisrt feature extraction module is used to receive acquired original voice data, and carries out feature to acquired original voice dataExtract；First normalization module is used to carry out the data after feature extraction Gaussian normalization processing, exports normalization data；
Neural network classification module is used to receive normalization data and the good deep neural network of training in advance, and by depthDegree neutral net carries out Classification and Identification to normalization data and obtains sorting result information, and sorting result information is acquired original soundThe no miscellaneous probable value containing impact sound of data, when the probable value exceedes threshold value set in advance, then judges exception occur in roadEvent；
Neural metwork training module is used to receive training data and the training of neutral net is carried out according to training data, obtainsAbundant learning tape is made an uproar the deep neural network of the Nonlinear Mapping relation between sample and clean sample, exports deep neural networkTo neural network classification module；
Neural metwork training module includes second feature extraction module, the second normalization module, unsupervised learning pre-trainingModule and supervised learning optimization module, wherein, second feature extraction module is complete with the functional structure of fisrt feature extraction moduleIt is identical, the feature for extracting training data；Second normalization module is identical with the functional structure of the first normalization module,Feature to being extracted carries out Gaussian normalization, i.e., into 0, it is 1 that variance is regular to the mean normalization of all training datas.It is unsupervisedStudy pre-training module carries out unsupervised learning initial training using training data is pre-processed as input, deep for initializing generationThe structure of neutral net is spent, the successively progressive learning neural network parameter by way of unsupervised learning, in finding input dataThe abstract characteristics of deep layer.Every layer of use of neutral net is restricted Boltzmann machine (RBM) model and carries out pre-training, by multipleRBM is superimposed as a depth network structure.Referring to Fig. 3, showing is restricted Boltzmann machine (RBM) structural representation, RBM'sSymmetrical connection is existed only between aobvious node layer and hidden node, and does not have any shape in the inside of aobvious node layer and hidden nodeThe connection of formula, it is believed that interlayer is full connection, is connectionless in layer.RBM is used as a kind of condition random field, each of which neuronNode describes a distribution situation for stochastic variable, and the higher order statistical phase in input vector is captured by each neuron nodeClosing property trains the potential rule included in input vector to explain and find.
Due to all not connected inside the aobvious layers of RBM and hidden layer, can very easily obtain each under data and model profileThe conditional expectation of state.For given training quantity v, the state of hidden node can be calculated by below equation (1)：
P(hi=1 | v)=σ (bj+∑viwij) (1)
Then use to sdpecific dispersion algorithm (CD1 algorithms, Contrastive Divergence, CD) to train RBM, then makeRBM parameters are updated with gradient descent algorithm：
In above formula (2), η is parameter renewal learning speed, and i is iterations.By the model parameter for adjusting RBM so thatReduced by the energy of the aobvious layer data specified of RBM, so that increase the probability that aobvious layer data occurs, and then training is arrived in RBM studyTrue distribution P (v) of data.
According to the method described above training complete a RBM after, study to weight fix, by training data calculateThe RBM hidden layer states for obtaining can be used to as the input data for training another RBM, namely train first using training dataIndividual RBM obtains a hidden layer L1 and its network weight W1, reuses the output of previous hidden layer as input data, successively instructsThe follow-up RBM of white silk obtains 2~W3 of hidden layer L2~L3 and network weight matrix W.Specific training process show this referring to Fig. 4The pre-training schematic diagram of RBM, all of network weight is successively initialized with this Greedy in invention, so that further unsupervisedStudy RBM Hidden units between dependence.After all of RBM has been trained, each RBM is superimposed, then mostLater layer is superimposed one softmax layers, thus constitute one it is bottom-up feedforward, deep layer, distinction for classifyingDeep-neural-network.Because the accumulation using RBM constitutes a depth network structure, having in this, as deep-neural-networkInitialization networking weight during supervised training, can prevent it to be absorbed in local optimum.
Trained by RBM generative natures successively, we can find a more preferable region in weight space, from thisRegion is set out, and the supervised learning of distinction can be made to optimize (intense adjustment) and proceed by supervision relative to from random initializtionPractise optimization (intense adjustment) and obtain better performance lifting, can also substantially reduce the possibility of over-fitting.Supervised in the present invention and learnedBackpropagation (back-propagation, BP) algorithm that optimization module is commonly used using prior art is practised, using labeled data pairNeural network parameter carries out the intense adjustment for having supervision.In the algorithm, two steps are generally divided into：1) response is propagated forward, will be defeatedEnter and exciter response is obtained by each hidden layer, and the output of last layer is next layer of input, is to the last predicted for one layerValue；2) backward error propagate, according to forward response travel to last layer, the prediction to signal can be obtained, this predicted value andThe difference of reference signal, exactly needs the mistake of backpropagation.Have the reversely mistake passed back, it is possible to according to this mistake comeAdjust each weight and the biasing of neutral net.It is ready to after the input data of DNN and output data, it is possible to start to updateThe weight and offset parameter of network, i.e. W and b, shown in equation below 3：
Here λ represents learning rate, and E represents an object function for optimizing, can be accurate using least mean-square errorThen；RepresentThe parameter for having weight to be learned and biasing of layer.L represents the number of hidden layer really, then L+1 is justRepresent output layer.By above-mentioned formula as can be seen that in the renewal process of model parameter, being almost set without any hypothesis,Therefore, DNN can well be fitted the non-linear relation that band is made an uproar between sample and clean sample.
In actual neural metwork training, it is the key factor for influenceing accuracy of detection that whether training data complete.In the present invention,Build " impact sound training dataset " and " road noise training dataset ", wherein, road noise training dataset is by realityThe voice data of various situations is gathered under the road environment of border and data are labeled；Impact sound training dataset collects various carsThe voice data of type impact test, and data are labeled according to impact strength；Clean impact sound collection is made an uproar with road respectivelySound is added together, obtains band and makes an uproar sample.By the above training data sample input neural network model training network weight andOffset parameter.Referring to Fig. 5, the structured flowchart of the deep neural network that training is obtained is shown, neutral net includes 1 inputLayer, L1~L3 and output layer of 3 hidden layers.During input signal feature extraction, signal is sampled 8KHz, corresponding everyIndividual frame length is set to 256 sample points (32 milliseconds), and it is 128 sample points that frame is moved, and Short Time Fourier Analysis are used to calculateThe DFT coefficient of each overlapping frame, therefore, input layer uses 128 nodes, and the dimension of correspondence input data, output layer is three-dimensionalData output, corresponds to pure noise, miscellaneous containing impact sound and miscellaneous containing voice respectively.Every node layer is 2048 in L1~L3, and it takesCertainly in the number of training data, 2048 correspondences, 1,000,000 training datas.The iteration of the pre-training of each limited Boltzmann machineNumber of times is 50 times, and the learning rate of pre-training is 0.0005, and the learning rate of the first two tuning for having supervision of ten times is 0.1, soLearning rate is successively decreased 10 every time afterwards, total iterations is 100 times.
The deep neural network trained using aforesaid way, with the increase of training data, systematic function is improved constantly,In class test, actual discrimination reaches 80%, the threshold value of early warning can be set into 60% in practice, can be used as road exceptionThe effective evaluation index of event early warning.
Referring to Fig. 6, the theory diagram of characteristic extracting module in the present invention is shown, fisrt feature extraction module/the second is specialLevy extraction module and further include framing module, DFT transform module and log power spectrum processing module, wherein, framing module is usedIn sub-frame processing is carried out to input data, using overlapping segmentation, the proportion that general frame shifting accounts for frame length is 0-50%；DFT transform mouldBlock obtains frequency domain information for carrying out discrete Fourier transform to the data after sub-frame processing；Log power spectrum processing module is used forLog power spectrum treatment is carried out to frequency domain information, the quadratic sum equivalent to each coefficient modulus after DFT transform is taken the logarithm, and it is right to takeNumber can simulate nonlinear perception characteristic of the human ear to the sound intensity, and information, than more complete, is not almost lost on log power spectrum in additionAny information is lost, is conducive to improving accuracy of detection.
In a preferred embodiment, impact sound identification module uses the artificial intelligence chip of build-in depths neutral netRealize.Although having the artificial intelligence chip of many powerful build-in depths neutral nets in the prior art, chip-scaleCalculating performance can't compare favourably with PC grades of calculating performance after all, cannot generally meet the requirement of real-time；Meanwhile, in this ShenIn applied environment please, car crass after all or small probability event (especially major traffic accidents), it is therefore not necessary to open in real timeNeural network classification module is opened to be identified.Referring to Fig. 7, impact sound identification module another embodiment in the present invention is shownTheory diagram, impact sound identification module also include average energy detection module and frame energy comparison module, wherein, average energyDetection module is used to calculate the average energy value of present frame log power spectrum and be sent to frame energy comparison module；Frame energy comparisonWhether module is used to calculate the difference of consecutive frame the average energy value and judges the difference beyond default threshold value, if it exceeds then openingOpen neural network classification module.Relative to the microphone that position is fixed, voice signal Energy distribution and the actual sound field of its collectionDistribution proportion relation.And in the applied environment of the application, under normal circumstances, the energy of voice signal is relatively flatSteady fluctuation, its fluctuation range can be previously obtained by practical measurement；And impact sound is a flash energy for burst, itsThe far super normal fluctuation range of intensity level.Therefore can decide whether pulse occur by front and rear frame energy comparison, if strong arteries and veinsRush energy to uprush, whether impact sound is (main to be then turned on flash during neural network classification module carries out judging this section audio signalDistinguish sound of blowing a whistle).Using above-mentioned technical proposal, unlocking condition is used as by frame energy anticipation, deep neural network is without realShi Yunhang, can reduce system power dissipation.
In a preferred embodiment, frame energy comparison module receives current information of vehicle flowrate, and it is flat to calculate present frameWhether equal energy value is beyond the energy value scope preset corresponding to corresponding information of vehicle flowrate, if it exceeds then opening neutral net pointGeneric module.Analyzed as above-mentioned, the noise that traffic current is produced is in proportionate relationship with vehicle flowrate, although various vehicle meetingsThere is different, be substantially opposite the energy range that specific vehicle flowrate has correspondence voice signal, we can be by adoptingVoice data in collection real road occasion is simultaneously analyzed to mass data this energy range is obtained ahead of time.Therefore, lead toCross judge current energy value whether correspondence vehicle flowrate energy range within as neural network classification module open stripPart, accuracy of detection can be further improved while neutral net operating time is reduced.
Referring to Fig. 8, show the theory diagram of sound acquisition module in the present invention, sound acquisition module further include byMicrophone array, audio processing modules and control module that multiple microphones are constituted, wherein, multiple microphones in microphone arrayIn certain geometrical shape and each microphone has unique ID；Audio processing modules are used to synchronously obtain and identify each wheatThe voice signal of gram elegance collection simultaneously exports audio-frequency information after processing voice signal；Control module and audio processing modules phaseConnection, the audio-frequency information for controlling the work of audio processing modules and after audio processing modules are processed is stored in memory moduleIn.Using above-mentioned technical proposal, by the audio signal of audio processing modules synchronous acquisition microphone array, the wheat of any one IDGram wind all gathers continuous audio-frequency information, and stores in a storage module.Due to setting multiple microphones, so as to ensure to be gatheredThe integrality of audio signal, compensate for the defect of audio quality difference when prior art video monitoring is applied in the road；Meanwhile, it is manyIndividual microphone is fixedly installed in certain geometrical shape, referring to Fig. 9, show the schematic diagram of microphone array arrangement, multiple uniqueThe mark rounded setting of microphone and synchronized sampling, in theory, when sound source sends sound in region, because sound source distance is everyIndividual microphone is apart from different, therefore each microphone receives the time that the intensity and signal of signal reaches and difference occursIt is different, therefore, the relative position information that the otherness signal received according to each microphone and each microphone determine just can be trueDetermine the location of sound source.
In order to determine the particular location of impact sound in time, referring to Figure 10, another side of being preferable to carry out of the invention is shownThe theory diagram of formula, also including impact sound locating module, impact sound locating module is connected with controller for road lamp, is touched for obtainingHit the particular location of sound and positional information is sent to controller for road lamp；When impact sound identification module judges anomalous event occurWhen, impact sound locating module obtains the temporal information of the anomalous event and corresponding microphone array is obtained from memory module and is listed in thisAudio-frequency information in temporal information, and according to each microphone fix position relationship and each microphone in the temporal informationThe parameter information of middle correspondence audio-frequency information determines the particular location residing for impact sound.Further, parameter information is each microphoneThe peak strength and each microphone of correspondence audio-frequency information are in the peak strength corresponding time difference in the temporal information.NamelyAfter impact sound identification module judges anomalous event occur in road, impact sound locating module can quickly determine anomalous eventPosition such that it is able to which work is dredged in timely and effective development rescue.
In a preferred embodiment, controller for road lamp is also connected with Rotatable camera device, rotatable shooting dressPut and be arranged on light pole and rotated according to the control instruction of controller for road lamp；In impact sound identification module judges roadWhen there is anomalous event, controller for road lamp control Rotatable camera device is rotated to specific position determined by impact sound locating modulePut.Using above-mentioned technical proposal, Rotatable camera device can more accurately collection site video, so as to pass through audio frequency and video knotClose, reduce the blind area of monitoring.
In the prior art, roadway lighting system is used to carry out Based Intelligent Control to each road lamp, so as to provide one comfortablyRoad lighting environment.Sound acquisition module can be integrated in road lamp, it is possible to reduce the quantities of system wiring.More preferablyGround, microphone array can be arranged in the lamp surface of road lamp, so that when existing road is transformed, without rewiring.
In a preferred embodiment, also including traffic flow detecting device, traffic flow detecting device is used to gather wagon flowAmount information is simultaneously sent to remote monitoring center or controller for road lamp.
In a preferred embodiment, the impact sound identification module uses the artificial intelligence of build-in depths neutral netChip is realized.Referring to Figure 11, the Organization Chart of the intelligent sound Processing with Neural Network chip CI1006 of present invention use is shown, beArtificial intelligent voice identification chip based on ASIC frameworks, contains deep neural network treatment hardware cell, perfect can prop upDNN computing frameworks are held, high performance data parallel is carried out, artificial intelligence deep learning voice technology pair is can greatly improveThe treatment effeciency of mass data；CI1006 can reduce dependence of the product for network using local Neural Network Data treatment,Lifting intelligent sound identification response and control speed.The chip since phonetic entry, speech detection, speech feature extraction andDNN computings are designed using hardware structure completely, and software is substantially carried out tone decoding and voice broadcast, compared to AP chip softwaresDNN schemes, with advantages such as operational performance and low cost, low-power consumption, small sizes higher.The chip can support local voiceDetection, wake-up, and the hundreds of identifications of offline order entry.Directly can also be replaced by the general controls interface of this chipThe original control MCU of equipment, realizes the voice-intelligent of equipment.The chip also has abundant Peripheral Interface, can pass throughThe external WIFI chips of the interfaces such as SPI, UART are connected to high in the clouds, after local wake-up, can be connected by high in the clouds and realize that nature is man-machineInteraction, or dock the types of applications service in high in the clouds.Off-network state is such as in, then automatically switches to local offline order word identificationFunction.
The explanation of above example is only intended to help and understands the method for the present invention and its core concept.It should be pointed out that rightFor those skilled in the art, under the premise without departing from the principles of the invention, the present invention can also be carried outSome improvement and modification, these are improved and modification is also fallen into the protection domain of the claims in the present invention.
The foregoing description of the disclosed embodiments, enables professional and technical personnel in the field to realize or uses the present invention.Various modifications to these embodiments will be apparent for those skilled in the art, as defined hereinGeneral Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, the present inventionThe embodiments shown herein is not intended to be limited to, and is to fit to and principles disclosed herein and features of novelty phase oneThe scope most wide for causing.