Target Tracking and Obstacle Avoidance for Mobile Robot based on Kinect

Transcription

1 3rd Internatonal Conference on Machnery, Materals and Informaton Technology Applcatons (ICMMITA 015) Target Trackng and Obstacle Avodance for Moble Robot based on Knect Mengxn LI 1, a,jad YIN, b 1 School of Informaton and Control Engneerng Shenyang Janzhu Unversty Shenyang, Chna School of Informaton and Control Engneerng Shenyang Janzhu Unversty Shenyang, Chna a lmengxn@sjzu.edu.cn b @qq.com Keywords:moble robot;target trackng ;obstacle avodance;knect;bfo-pf;artfcal potental feld; Abstract.In order to get a better understandng of the obstacle avodance of a moble robot when t s trackng a target,we used Knect to take the place of the tradtonal range radar and camera.trackng part uses the mproved partcle flter algorthm.to deal wth the partcle degeneracy problem, a new partcle flter (PF) based on the Bactera Foragng Optmzaton (BFO) algorthm s proposed.improved artfcal potental feld method ncreases the safe area whch was appled to remove the nformaton of an obstacle not dsturbng the moton of the robot,so as to mprove the ablty of the moble robot to pass through a narrow passage by the tradtonal artfcal potental feld method.the results show that the system can accomplsh the task of trackng and obstacle avodance. Introducton Ultrasonc and laser rangng radar s the man sensor of the moble robot to obtan envronmental nformaton. However, the ultrasonc radar to obtan the envronmental nformaton s more rough, laser radar, although the acquston of nformaton s rch, but the cost s very hgh. Knect s a 3-D camera produced by the Mcrosoft Corp. It can obtan the poston nformaton of the envronment optcal mage and optcal mage. The nformaton s rch and low cost. It s sutable for moble robot path plannng and target trackng. In ths paper, the target trackng and obstacle avodance system of moble robot s constructed by usng Knect nstead of tradtonal rangng sensor.combned wth the mproved artfcal potental feld obstacle avodance algorthm[1-4] and partcle fler (PF) based on the Bactera Foragng Optmzaton (BFO) algorthm[5-7], the target trackng s realzed on the bass of the dfferental drven wheeled moble robot. Moble Robot Wth Knect Moble Robot Knematc Model At present, the moble robot s used for the wheel dfferental drve, whch s descrbed n the nertal coordnate system. x cos 0 v cos. v y sn 0 w v sn. 0 1 w (1) where: x,y coordnates of the moble robot n the global reference frame, s the drecton of the moble robot n the global reference frame angle, v s the speed of the moble robot moves lnearly, 015. The authors - Publshed by Atlants Press 1116

2 s the angular velocty of rotaton of the moble robot. Fgure 1 s a moble robot nertal coordnate system, where Xr, Yr determne the local coordnate system of the moble robot.x,y establshed ts world coordnates. Fg.1 Inertal coordnate system of the moble robot Accordng to correspondence between the moble robot global coordnate system and local coordnate system, the local coordnate system can transform coordnates relatve to the global coordnate system of coordnates. Order the coordnates x l, y l to coordnate robot local coordnate system, and t s known n the global moble robot coordnate system poston xc, y c, c, then the relatve coordnates global robot coordnate system x, by equaton (1). xg xc xl cosc ylsn c, () yg yc xlsnc yl cos c. Knect Mcrosoft launched Knect somatosensory camera one for XBOX360, currently supports the Knect software development kt for the use of personal computers and Mcrosoft have OpenNI provded SDK, Knect va USB Interface connected to the computer. Use Knect can get a resoluton RGB color mages was 640*480and a correspondng depth mage. The use of the depth nformaton obtaned, can be obtaned by convertng the correspondng 3- D model of the envronment, the converted three-dmensonal coordnate system shown n Fgure, the orgn O of the camera locaton, Z-axs postve drecton n front of the camera. g y g Fg. Knect and mage coordnate system Snce dfferent coordnate systems coordnate system Knect camera and moble robots, so 3- D coordnate obtaned from Knect need to go through the converson n order to ensure consstency wth the moble robot coordnate system. Specfc converson relatonshp shown n Fg.3, the fgure s the Knect for 3- D X,Y,Z coordnate system, Xr, Yr s local coordnate system for the moble robot. Fg.3 Coordnates relatonshp between Knect and moble robot Target Trackng and Obstacle Avodance Partcle Flter Algorthm Based On Bacteral Foragng Optmzaton 1) The core of the bacteral foragng algorthm s the trend of the nner layer. Generally, trend 1117

3 operaton s just a random search method.the method avods the blndness of random search and mproves the searchng effcency. It s gven by the followng. best ( j) sgn( ) (3) ( j 1, k, l) ( j, k, l) ( C( ) rand()) ( j) (4) best The proposed method (3) gves the mproved search drecton, whch represents the poston of the current group of ndvduals, and sgn represents the symbolc functon; (4) gves the fnal trend operatng formula, whch rand() means that a random number U [0,1] of the dstrbuton, the ntroducton of random step sze factor can expand the search area and avod the condton that the fxed step sze s lkely to mss the optmal search results. ) Mgraton operaton s the trend of the operaton, ts essence s the mgraton of ndvduals wth a certan probablty. chromosome mutaton, whch s smlar to the genetc algorthm n the lterature therefore, mutaton method proposed by [8] for the basc mgraton operaton s mproved. ( j, kl, 1) ( j, kl, ) c (5) In the upper formula, c s a Gauss random varable, c~ N( y( t), q) s subject to the observaton of the current tme, and q s defned as a constant of the varance. 3)Smulaton And Analyss Accordng to the lterature [9,10,11] state equaton and observaton equaton usng non-statonary growth model, the model by as follows. 5xk 1 xk 0.5xk 1 8cos[1.( k1)] u k1 1 xk 1 xk yk vk 0 (6) The above model,state nose uk ~ N (0,10),observaton nose vk ~ N (0,0.1),both are ndependent.the ntal state of the system s set to 0.1. Respectvely wth SPF, GA-PF, APF, and BFO-PF algorthm the model proposed state estmaton. Bacteral foragng algorthm parameters set, Nc 4, r Genetc parameter s set, N 0, P 0.6 P 0.4, cross mutaton N N, P 0.8, The number of partcles s set to 100, Fg. 4shows the smulaton results of a sngle experment TRUE SPF APF GA-PF BFO-PF ed ed 5 估计值 时间 Fg. 4 Trackng performance of dfferent algorthms As t can be seen from Fg. 4, compared wth other algorthms, BFO-PF algorthm n real value trackng process, closer to the real value. To further verfy the performance of the algorthm, usng the root mean square error of several Monte Carlo experments (RMSE) as an evaluaton ndex, root mean square error s defned multple Monte Carlo experment s gven by the followng equaton. 1 RMSE ( ) MC MC xk x k (7) k 1 where, MC for the Monte Carlo trals, xk s the state of true value, xk the state estmated value. Partcle number s set to 100, the monte carlo experments for 400. Table 1 shows the RMSE of 1118

4 mean and varance of dfferent algorthm, and the average elapsed tme. TABLE I. RMSE mean and varance of dfferent algorthms and the average runnng tme Partcle flterng method RMSE Mean RMSE Varance Average runnng tme SPF APF GA-PF BFO-PF From table 1,t can be seen that the BFO-PF method proposed n ths paper s superor to APF and SPF algorthm n flterng performance.it shows the effcency of the algorthm. Obstacle Avodance 1)Securty zone artfcal potental feld Obstacle avodance of moble robot usng artfcal potental feld s completed, the method s smple and practcal and clear physcal sgnfcance and thus subject to extensve research and use. Specfc formula for artfcal potental feld.in order to acheve the goal of target trackng and obstacle avodance between swtchng desgn wth regonal securty of the artfcal potental feld method, mplementaton methods, such as Fg.5. Fg.5(a) a sold rectangle frame representaton of the normal runnng of regonal securty, Fg.5(b) a sold rectangle box have obstacle 3, at ths pont the system preferentally nto the obstacle avodance mode, moble robot accordng to the dotted rectangle frame determned wthn the scope of the obstacle nformaton obstacle avodance untl the small box there s no obstacle. Ths method can not only effectvely control mode swtchng, but also enhances the whole system through the narrow channel and the ablty to walk along the wall. (a)no obstacle (b)obstacle Fg.5 The safe regon of artfcal potental )Expermental results and analyss The experment used a length of 900 mm wdth s about 660 mm heght 480 mm moble robot usng a dfferental drve mode, and no turnng radus lmt. Insde ths robot was equpped wth a computer, you can easly connect wth Knect.The expermental results shown n Fg.6, shown n red dots ndcate the moble robot trajectory, long black bars represent obstacles, rectangle shows a moble robot. Moble robot eventually stops movng target at a dstance of about d, set D n the experment s 1m. As can be seen from Fg. 6 moble robots n trackng targets whle movng to effectvely avod obstacles. Fg.6 The expermental results 1119

5 Concluson Expermental results show that the use of Knect may well complete moble robot target trackng and obstacle avodance task, we propose a bacteral foragng optmzaton based on partcle flter resamplng mproved partcle depleton problem exsts partcle flter. In ths paper, two-dmensonal model, n future studes, f the three-dmensonal model nto the algorthm, moble robot wll be able to better track and obstacle avodance. Acknowledgement In ths paper, the research was sponsored by the Nature Scence Foundaton of Chna (Project No ) References [1]GAO Feng,WANG Jangfeng,SHI Shaoyou,et al. Vehcle ntellgent collson warnng algorthm based on fuzzy neural networks[j].journal of Jangsu Unversty: Nature Scence Edton,006,7( 3): []WANG We, WANG Hua, School of Astronautcs. Real-tme obstacle avodance trajectory plannng for mssle borne ar vehcle based on constraned artfcal potental feld method[j]. Journal of Aerospace Power, 014. [3]Lu Z Z. The Applcaton of Artfcal Potental Feld n Intellgent Unmanned Vehcle Obstacle Avodance System[J]. Computng Technology & Automaton, 013. [4]Yang Y B, Wang C L. Obstacle Avodance Method for Moble Robots Based on Improved Artfcal Potental Feld Method and Its Implementaton on MATLAB[J]. Journal of Unversty of Shangha for Scence & Technology, 013. [5]YangXn, LuJa, Pengyu Zhou, etc. Based on feature fuson of adaptve partcle flter target trackng algorthm. Journal of Jln unversty (engneerng scence), 015, :09. [6]Ruj Zhu, Naxn Q, shengxu Zhang, et al. A knd of mproved nfrared target trackng algorthm based on partcle flter. Journal of Schuan armaments factores, 014, 35 (3) : [7]XIA Xuan,LIU Huapng.DSP- based actve vson system[j].robot,01,34( 3): [8]PARK S., WANG J.H., ROU K., KIM E. A New Partcle Flter Inspred by Bologcal Evoluton: Genetc Flter. World Academy of Scence, Engneerng and Technology 33, 007 [9] Le Lang, Ll Zhao. Based on artfcal fsh swarm algorthm (J). The computer s appled to software, 01,9 (1): [10] Zhyu Zhu. Partcle flter algorthm and ts applcaton. Scence Press, 010: [11]ARULAMPALAM M. S., MASKELL S., GORDON N., CLAPP T. A Tutoral on Partcle Flters for Onlne Nonlner/Non-Gaussan Bayesan Trackng. IEEE Trans. on Sgnal Processng, Vol.50 No. February