A kind of unmanned security system, method and sweeping robotTechnical field
The present invention relates to Intelligent robot for sweeping floor fields, and in particular to a kind of unmanned security system, method and machine of sweeping the floorPeople.
Background technology
When it is indoor nobody when be easy to happen the criminal cases such as burglary, larger economic loss is caused to resident family, evenIt is possible that can be because resident family returns in home, resident family and criminal surprisingly meet and cause damages to resident family's personal safety temporarily.ApplicationNumber be 201720993039.7 Chinese patent, disclose a kind of household robot with security function, including fuselage, bottomDisk, the first support plate, the second support plate, steering motor, threaded rod, shaft, driving gear, U-shaped frame, deflecting roller, driving device,It is motor, turntable, camera, infrared sensor, headlamp, Fresnel Lenses, telephone connection net alarm, control panel, wirelessCommunication module, GPS positioning module, PLC controller, annular protrusion, annular groove and reception terminal, the bottom of fuselage is equipped with bottomDisk is fixed with the first support plate, the second support plate and steering motor at the top of chassis, between the first support plate and the second support plateIt is provided with threaded rod, the unsupervised to house security and guard is realized, home room can be patrolled by preset pathMonitoring is patrolled, remote household member can see patrol picture at any time by receiving terminal, once someone swarms into, the household robotMeeting automatic alarm, ensure that the safety of family.
Sweeping robot also known as sweeps machine, intelligent dust suction, robot cleaner etc., is the one of controlling intelligent household appliances automaticallyKind, certain artificial intelligence can be relied on, complete floor cleaning work in the room automatically.General use brush is swept and vacuum mode,Ground sundries is first received into the rubbish storage box into itself, to complete the function of land clearing.In general, it will complete clearSweep, dust suction, the robot to work to wiping, be also uniformly classified as sweeping robot, existing sweeping robot only has wipe cleanFunction, have a single function, intelligence degree is low, and the fun of using is few, and robot use environment mostly home environment, therefore oneHome security systems assistant officer generation exploitation of the kind based on sweeping robot.
Invention content
To solve the above problems, the present invention provides a kind of unmanned security system, method and sweeping robot.Base of the present inventionIn the intrinsic patrol function of sweeping robot and path planning function, video input function is swept the floor using software system developmentThere is the peripheral applications of robot portability, user to be provided by family of low cost without individually purchase safety protection robotSecurity system, avoids the wasting of resources.
To realize the technical purpose, the technical scheme is that：A kind of unmanned security system, including video input listMember, image processing unit, Moving target detection unit, mobile target classification unit, character features extraction unit, character featuresWith unit, alarm unit；
The video input unit is used to obtain the ambient video where robot, and described image processing unit is used for instituteThe video for stating the input of video input unit carries out RGB data processing by frame；
The Moving target detection unit is used to, according to RGB data treated video, carry out RGB information comparison frame by frame,Judge mobile target；The mobile target classification unit distinguishes people and pet for classifying to mobile target；The peopleObject feature extraction unit and for character features matching unit be respectively used to character features extraction and matching, realize person recognition,Judge that stranger invades；The alarm unit after in character features matching unit, it fails to match for alarming.
Further, the mobile target classification unit uses neural network model comprising：Input layer, Standard convolution layer,Separable convolution packet, averagely drop latitude layer are fully connected layer, output layer；The input layer carries out RGB to described image processing moduleVideo after data processing is read by frame；The output layer is used to export the pet model and mock-up funiture, Yi Jishi in videoThe confidence level of other result.
Further, further include timing unit, mobile patrol unit；
Mobile patrol unit, is used for mobile robot, environment carries out patrol pattern indoors；
Timing unit is used for according to artificial setting, and in special time period, time opening goes on patrol pattern.
Preferably, further include sound detection unit, for detecting indoor sound by sound transducer, pass through the movementGo on patrol unit so that robot carries out patrol pattern.
A kind of unmanned security method, is utilized above-mentioned unmanned security system, includes the following steps：
S1：When robot is gone on patrol, the video in camera visual range in robot is recorded, and video is carried out at imageReason；
S2：Rgb color, Pixel Information extraction are carried out frame by frame to the video after image procossing in the step S1 and compared,Calculate mobile objective contour；
S3：Mobile objective contour in the step S2 is classified, it is people or pet to distinguish mobile target, such asFruit is that pet then carries out step S1, otherwise carries out step S4；
S4：Extract the character features that mobile target in the step S3 is people；
S5：Character features in the step S4 are matched with the user characteristics of sweeper robot Historic preservation,It alarms if it fails to match, otherwise carries out the step S1.
Further, the method that mobile target is people or pet is distinguished in the step S3, using including input layer, markQuasi- convolutional layer, the neural network model for separable convolution packet, averagely dropping latitude layer, being fully connected layer, furniture identification output layer, to instituteIt states image processing module to carry out RGB data treated that video is read by frame, by carrying out Standard convolution to every frame image, can divideIt from convolution, averagely drops latitude, be fully connected processing, judge the furniture in image, the mock-up funiture in output video and identificationAs a result confidence level.
Further, the method for the separable convolution includes depth convolution operation, line rectification operation, batch standardization behaviourMake and point-by-point convolution operation.
Further, further include unmanned security mode setting step before the step S1：
T1：The unmanned security mode time section of user setting opens unmanned peace in unmanned security mode time section inner machine peopleGuarantor's pattern；
T2：In step T1 described in user setting in unmanned security mode time section, the timing at robot patrol interval,After timing, the step S1 is carried out.
As a kind of preferred embodiment of the present invention, it is based on the above, unlike：
Before timing in the step T2 terminates, indoor sound is detected by sound transducer, is had indoorsThe step S1 is carried out when quiet.
A kind of sweeping robot, including above-mentioned unmanned security system and unmanned security method.
The beneficial effects of the present invention are：
The present invention passes through image processing techniques and nerve by going on patrol the real-time video of shooting home environment to robotNetwork technology judges, has distinguished mobile object, and when mobile object is behaved, and carries out person recognition, illegally enters to realizeThe differentiation invaded；Meanwhile the present invention has also set up a kind of patrol mechanism, substantially increases resource utilization and security performance；And thisInvention utilizes software system development based on the intrinsic patrol function of sweeping robot and path planning function, video input functionThe peripheral applications of sweeping robot have portability, and user is without individually purchase safety protection robot, using low cost as familyFront yard provides security system, avoids the wasting of resources.
Description of the drawings
Fig. 1 is the module principle figure of the unmanned security system of the present invention；
Fig. 2 is the mobile target classification unit schematic diagram in Fig. 1 of the present invention；
Fig. 3 is the flow chart of the unmanned security method of the present invention；
Fig. 4 is the flow chart of the separable convolution algorithm of the present invention.
Specific implementation mode
Technical scheme of the present invention will be clearly and completely described below.
As shown in Figure 1, a kind of unmanned security system, including video input unit, image processing unit, Moving target detectionUnit, mobile target classification unit, character features extraction unit, character features matching unit, alarm unit；
The video input unit is used to obtain the ambient video where robot, and described image processing unit is used for instituteThe video for stating the input of video input unit carries out RGB data processing by frame；
The Moving target detection unit is used to, according to RGB data treated video, carry out RGB information comparison frame by frame,Judge mobile target；The mobile target classification unit distinguishes people and pet for classifying to mobile target；The peopleObject feature extraction unit and for character features matching unit be respectively used to character features extraction and matching, realize person recognition,Judge that stranger invades；The alarm unit after in character features matching unit, it fails to match for alarming.The present invention is logicalThe real-time video for going on patrol shooting home environment to robot is crossed, and by image processing techniques and nerual network technique, is judged, areaMobile object is divided, and when mobile object is behaved, person recognition has been carried out, to realize the differentiation of illegal invasion.
Further, as shown in Fig. 2, the mobile target classification unit uses neural network model comprising input layer(input), Standard convolution layer (convl), separable convolution packet (module1-9), averagely drop latitude layer (average pool), completeFull articulamentum (full connect), output layer (output)；The input layer carries out RGB data to described image processing moduleTreated, and video is read by frame；The output layer is used to export pet model in video and mock-up funiture and identification knotThe confidence level of fruit.
Further, further include timing unit, mobile patrol unit；
Mobile patrol unit, is used for mobile robot, environment carries out patrol pattern indoors；
Timing unit is used for according to artificial setting, and in special time period, time opening goes on patrol pattern.
Preferably, further include sound detection unit, for detecting indoor sound by sound transducer, pass through the movementGo on patrol unit so that robot carries out patrol pattern.Sound transducer functions as a microphone (microphone).It is used forSound wave is received, shows the vibrational diagram of sound, but cannot be measured to the intensity of noise.Built in sound transducer one to soundThe condenser type Electret condenser microphone of sound sensitivity, sound wave make the electret film in microphone vibrate, and lead to the variation of capacitance, and generate withCorrespondence variation small voltage, this voltage is subsequently converted into the voltage of 0-5V, converts by A/D and is connect by data collectorBy, and send processor to and realize sound induction, this is the conventional technical means of those skilled in the art, and this will not be repeated here.
As shown in figure 3, a kind of unmanned security method, is utilized above-mentioned unmanned security system, includes the following steps：
S1：When robot is gone on patrol, the video in camera visual range in robot is recorded, and video is carried out at imageReason；
S2：Rgb color, Pixel Information extraction are carried out frame by frame to the video after image procossing in the step S1 and compared,Calculate mobile objective contour；
S3：Mobile objective contour in the step S2 is classified, it is people or pet to distinguish mobile target, such asFruit is that pet then carries out step S1, otherwise carries out step S4；
S4：Extract the character features that mobile target in the step S3 is people；This character features can be based on it is above-mentionedThe facial character features of rgb color Pixel Information can also be the 3D body characteristics etc. established using depth camera.
S5：Character features in the step S4 are matched with the user characteristics of sweeper robot Historic preservation,It alarms if it fails to match, otherwise carries out the step S1.The mode of this alarm can be short message, app prompting messagesEtc. modes.It should be noted that the present invention can also pass through face recognition technology, shape recognition technology, sound groove recognition technology in e etc.Multiple technologies means carry out person recognition.
Further, the method that mobile target is people or pet is distinguished in the step S3, using including input layer, markQuasi- convolutional layer, the neural network model for separable convolution packet, averagely dropping latitude layer, being fully connected layer, furniture identification output layer, to instituteIt states image processing module to carry out RGB data treated that video is read by frame, by carrying out Standard convolution to every frame image, can divideIt from convolution, averagely drops latitude, be fully connected processing, judge the furniture in image, the mock-up funiture in output video and identificationAs a result confidence level.Wherein this neural network model can early period carry out artificial top set data set collect with mark pet withPeople, later stage are based on MobileNet frameworks progress disaggregated model pre-training and obtain.
Further, as shown in figure 4, the method for the separable convolution includes depth convolution (depthwise conv) behaviourMake, line rectification (RELU) operation, batch standardizes (batch norm) operation and point-by-point convolution (1 × 1conv) operation.
Further, further include unmanned security mode setting step before the step S1：
T1：The unmanned security mode time section of user setting opens unmanned peace in unmanned security mode time section inner machine peopleGuarantor's pattern；This period can be the time of having a rest in the evening, can also be to go out the time on daytime.
T2：In step T1 described in user setting in unmanned security mode time section, the timing at robot patrol interval,After timing, the step S1 is carried out.That is, the present invention establishes a kind of patrol mechanism, this mechanism canTo be to open unmanned security pattern, and in unmanned security pattern, machine is artificially spaced progress based on the period manually setSecurity is gone on patrol, while being spaced in patrol and opening sound detection, when judging to have sound, patrol security is carried out in time, greatly improvesResource utilization and security performance.
As a kind of preferred embodiment of the present invention, it is based on the above, unlike：
Before timing in the step T2 terminates, indoor sound is detected by sound transducer, is had indoorsThe step S1 is carried out when quiet.
A kind of sweeping robot, including above-mentioned unmanned security system and unmanned security method.Advantage is, is based onThe intrinsic patrol function of sweeping robot and path planning function, video input function utilize software system development sweeperThere is the peripheral applications of device people portability, user to be provided by family of low cost without individually purchase safety protection robotSecurity system avoids the wasting of resources.
For those of ordinary skill in the art, without departing from the concept of the premise of the invention, it can also doGo out several modifications and improvements, these are all within the scope of protection of the present invention.