GPS-less animal tracking system

GPS-Less Animal Tracking System Apurva Joshi, Naga VishnuKanth I, Navkar Samdaria, Sumit Bagla and #Prabhat Ranjan Dhirubhai Ambani Institute of Information and Communication Technology Near Indroda Circle, Gandhinagar, Gujarat, India 382007 {apurva_joshi,vishnu_kanth,navkar_samdaria,sumit_bagla,prabhat_ranjan}@daiict.ac.in

Abstract—Mobile target tracking in Sensor Network field is an important area of Wireless Sensor Network (WSN) application. The speed of target and constraints on WSN nodes vary from one application to another. In this paper, we report our work on application of WSN to track movements of Small Turtles in Wildlife Institute of India (WII) Campus. This would allow wildlife researchers to understand behaviour of these turtles. These turtles are very small in size and Global Positioning System (GPS) based monitoring is not very useful. So we have developed a GPS-less system to track movement of these turtles. In addition, we monitor the micro-climate around the animal as well as monitor period of hibernation during winter time. We have carried out trial of this system in our campus with a simulated mobile target carrying one node with a grid of four fixed sensor nodes. This system uses Incremental Grid Based Approach to localize moving target. The data obtained is passed on to a web based data server through a gateway and movement is shown on a digital map using a browser based visualization software utilizing Google Map APIs (Application Programming Interface). Keywords: GPS-less tracking, Localization, Mobile target tracking, Sensor Network, Wildlife tracking.

I. INTRODUCTION We have been working on a number of applications of WSN to help wildlife researchers. They cover a variety of animals of differing size and behaviour. In this paper we report on the work done on monitoring the movement of small animals (e.g. small turtles) in WII, Dehradun campus through a network of sensor nodes. The sensor network permits the monitoring of movement patterns, micro-climatic preferences and hibernation details of turtles. These turtles mostly inhabit bushy areas in the WII campus. They move around in search of food and during winter time go into hibernation. Traditionally they have been studied by placing an identification mark on their shell and recording where they have been found. Later on if they are found again, their location and time is noted. This method mostly depends on the possibility of a human locator being able to locate it and between two readings a lot of time elapses. We do not get any information about their behaviour during this time. We are trying to apply WSN technique of locating a mobile target to small animals such as turtles. This would allow wildlife researchers continuous information about the movement of animals within the campus. Since these animals

move at very slow speed and remain confined to a limited region within the campus, WSN based tracking is a good possibility. We also try to provide information regarding micro-climate by placing temperature, light and relative humidity sensor on the node being placed on the body of animal. We also place an accelerometer on the body to detect when animal goes into hibernation. All this information on continuous basis would provide researchers with a wealth of information compared to what has been possible traditionally. We have described basic methodology in [1]. However we briefly describe some of the aspects here for continuity. One of the techniques used for wildlife tracking is placing GPS on each mobile sensor node, which becomes expensive in terms of cost, energy consumption and also increases the size and weight of the sensor node. Use of GPS to track small animals like turtles will exceed the maximum permissible limit of about 25 gms for a device which can be placed on the body of small turtles (typically weighing less than 500 gms). Thus we need to use a localization technique which does not depend on GPS technology. So we need to deploy a GPS-less technology for tracking such small animals. We describe the methodology, we have adopted for this system in next section. II. METHODOLOGY We plan to set up a grid of receiving stations carrying fixed sensor nodes (anchor nodes). Currently we are assuming that these would be deployed in a regular manner assuming uniform 2-dimensional terrain. A mobile node (animal node) would be deployed on the body of the turtles to be tracked. The mobile node would have a radio transceiver and would send out radio beacons in form of a data packet periodically. The data packets would contain a sequence number to identify the packet and may contain sensor readings as well. These beacons would be received by one hop anchor nodes and they would record signal strength of the packet in form of RSSI value as well as link quality. Since the system would be deployed within the WII campus, we are assuming that anchor nodes would be ON all the time and listening to incoming packets. This is possible due to the fact that a number of street light posts are available near the area of deployment. This removes one of the constraints of WSN i.e. ON/OFF duty cycling the nodes to reduce power consumption and make the task at hand easier to handle. However animal node would be limited in its energy resource and would need lot of effort to keep the energy consumption to be as low as possible without

exceeding the weight limit and performing the desired functionality without reducing the life time. This becomes the main challenge for developing such nodes. We plan to do this by selecting low power consuming components as well as duty cycling the animal node. Reduction of communication traffic would be achieved by local processing of information as much as possible. In absence of any direct location sensors, we would need to utilize indirect way of estimating location of the animal node. We have discussed the various possibilities of it in [1] and also described localization algorithm we plan to use. While there may be possibility of using some other algorithms, right now we would use only one of them to make sure that rest of the system works reliably before we try other algorithms. As mentioned earlier, our localization algorithm is based on the Received Signal Strength Indicator (RSSI), which is discussed in detail in the following section.

value and is incremented every time it falls in the region overlapped by the set of rings. Finally the grid cells with the maximum count are identified and their centroid is used as the estimated location of the animal node.

Figure 2: Incremental grid based algorithm

III. LOCALIZATION ALGORITHM In this section we discuss an “Incremental Grid Based Approach” [2], which is based on measurements of received signal strength. Consider four anchor (fixed) nodes A, B, C and D placed at the corner of the square area, such that the animal node (mobile) is in the transmission range of the four anchor nodes. We assume that the positions of the anchor nodes are known to us (either by placing them at known position or by using GPS). The square area enclosed by the four anchor nodes is defined as grid. The grid is divided into large number of grid cells. Each side of the grid is calibrated with the readings of RSSI values measured by the incoming animal node packet, taken along each side of the square in both directions of the anchor nodes forming an RSSI matrix. The complete algorithm is based on this RSSI matrix which can be termed as “Environmental matrix” or “Calibration matrix”. This environmental matrix is dependent on various factors like atmospheric condition, nature of terrain, density of vegetation, etc. The animal node periodically transmits a data packet as a beacon. The anchor nodes receive the signal and record its RSSI value and send this value to the central system or the base station. Based on the RSSI reading taken earlier along the sides of the square (environmental matrix), adjacent grid points are identified which bound the measured value.

Figure1: Four anchor nodes each at corner of the square

From each of these grid points we draw rings with centre as the respective anchor node. So for each anchor node we have certain set of rings. Each grid cell is initialized with a zero

For example, in figure 2, we have a square area with four anchor nodes A, B, C, and D at its four corners. RSSI readings are taken along each side of the square in both directions to-and-from the anchor nodes. The readings are taken at regular intervals of distance. From A, the readings are a2,1 , a2,2 and so on up to a2,n towards B. Similarly from A towards D, the readings are a1,1 , a1,2 and so on up to a1,n. For each beacon packet, base station would collect a set of RSSI values received by four anchor nodes with same sequence number. Base station would carry out the following tasks to estimate the location: 1. Starts with anchor node A. First compares the RSSI values with measured signal strength along the sides connecting the anchor node A to its neighbors. It marks the intervals between which the RSSI values lies. For example in figure 2 it lies between a1,2 and a1,3 in the horizontal direction and a2,2, and a2,3 in the vertical direction. It is possible that we may get multiple intervals for some values. We keep all the intervals for our purpose. 2. Now from each interval we draw a ring covering the grid cells. 3. Grid cell count of the cells falling in between the rings is incremented, implying a stronger possibility of the animal node lying within that region. 4. It repeats the procedure 1 to 3 for the rest of the received RSSI values obtained from the 3 remaining anchors. 5. The centroid of the cells with the maximum count gives us the approximated location of the mobile node. It is easy to see that the size of the grids determines the possible smallest granularity of location error. Small grids are thus preferred but small grids need more calculation time and more calibration effort.

IV. SYSTEM ARCHITECTURE In applications like wildlife monitoring, nodes may have different capabilities and execute different functions. For example, some nodes may have larger battery capacity and more powerful processing capability; which would allow them to serve as data gathering nodes (gateways). Some nodes may only be responsible for sensing and sending collected data to the gateway nodes. In GPS less tracking of turtles, the three types of nodes corresponding to the system being implemented for the project are: 1) Animal Node (mobile) - These individual sensor nodes are affixed to each of the turtles being tracked. 2) Anchor Nodes – These nodes propagate the signals from the animal node to the base station 3) Base Station - An onsite high power station which relays collected data back to the data server via a suitable user interface. Each type of node has a different requirement. Sensor nodes are built using XBee and XBeePro Modules by MaxStream. The XBee and XBeePro OEM RF Modules support the unique needs of low-cost and low-power wireless sensor networks. These modules require minimal power and provide reliable delivery of data between devices. The modules operate within the ISM 2.4 GHz frequency band and are pin-for-pin compatible with each other. The hardware used for animal node consists of an XBee module. Since the animal node has to last long without any intervention, there should be an efficient utilization of power. XBee being low on power consumption is therefore used in animal node. Whereas XBeePro, due to its larger transmitted power, is used in anchor nodes that are required to communicate over longer distances. As the base station can be powered with high capacity, it is provided with a more range efficient XBee Pro module. The XBeePro module improves the receiving capacity of the base station. Along with XBee modules, animal and anchor nodes also have an Atmega32 microcontroller [3] and a 3.3V voltage regulator (sensors will be annexed at later stage of the project).

Figure 3: System Architecture

As shown in the figure 3, it can be noticed that, the overall system would be viewed as a grid with four anchor nodes placed at the corresponding corners which continuously relay the data received from the animal node to the central server or the base station. Dotted arrow lines in the figure 3 between the two nodes indicate that the two nodes are in the transmission range. In case one of the anchor nodes is not

able to send data directly to base station, it would relay it through another anchor node. V. COMMUNICATION PROTOCOL As the communication between nodes is being carried out wirelessly, it is essential to follow a suitable and robust communication protocol for effective transmission of data between nodes. The proposed protocol depends on the type of node transmitting the data. Each XBee of a node is assigned a unique 16-bit address, thereby making each node unique. XBee are made to function in API mode [4]. API mode enables transmission of data in the form of packets called frame. Each frame transmitted by a node is appended by its node ID ensuring the receiver to identify the node from which the data is being received. The communication takes place in two steps: 1) Between the animal node and the anchor node. 2) Between the anchor node and the central base station. Animal node is programmed to broadcast burst of ten packets having same sequence number at every interval of 5 minutes. Each packet contains sequence number, animal node ID and additional relevant data (data from sensors regarding temperature, humidity etc.). Each burst of packets bear a unique sequence number. The packets from the animal node are then broadcast (multicast) to the anchor nodes. Anchor nodes receive the packets from animal nodes. Anchor nodes use inbuilt functionality of XBee to find RSSI value and then calculate its average value for each animal node based on common sequence number. Anchor node creates a new packet comprising of average RSSI value, anchor node ID, animal node ID and sequence number of the corresponding received packet. The data is then transmitted to base station using base station address (unicast mode). VI. IMPLEMENTATION The system implementation requires a grid network to be formed at the area where turtle has to be tracked. The grid consists of an anchor node at each of its corners. The node implanted on turtle’s shell communicates with anchor nodes present in its vicinity. The anchor nodes in turn communicate the data received from the animal node to the base station. The data is stored at the base station. The programs running at the base station continuously processes the data when complete set of data for a sequence number is received. It then calculates the distance and bearing angle of the mobile node with respect to the top right anchor node, ‘D’. The records whose data set is not complete are kept in buffer and they are processed as and when the rest of the data corresponding to the same sequence number arrives. The records, for which the data set is complete, are analysed and the program generates a matrix representing the count of the grid cells. Figure 4 shows the state of the matrix when the algorithm tries to determine the most probable location of the mobile node using the signal strength received

at the anchor node located at the top right corner of the figure 2. Each number represents the count of the smaller cell. The arc made by the non-zero numbers is distinctly visible. Moreover, since the localization is done based on experimental RSSI values with respect to distance, multiple rings are formed for the analysis of a single node. Therefore, the count of the smaller cells in the diagram can be noted to be more than 1 in many cells. The smaller cells having higher counts are surrounded by the smaller cells having comparatively lesser counts.

Figure 6: Testing location with four anchor nodes

Figure 4

VII. FIELD TRIALS Trials were conducted at the campus of Dhirubhai Ambani Institute of Information and Communication Technology (DA-IICT), Gandhinagar, Gujarat. A set of preliminary tests under same environmental conditions were conducted for measuring the variation on signal strength (RSSI value) with distance for a RF transceiver. It was found that averaging multiple measurements over time can give more precise location as compared to taking single measurements over a given period of time. Experimental data for variation of RSSI with distance is shown in figure 7. A calculated value of signal strength based on free space propagation (calculated value) is also shown along with the measured value (with average and without average).

Figure 5

The final matrix as shown in the figure 5 is the addition of the four matrices obtained by incrementing this matrix with respect to each anchor node. The probable location of the mobile node is taken to be the centroid of the cells with the maximum count (which in this case is 8). The distance and bearing angle of the mobile nodes with respect to the anchor nodes is then used to find the latitude and longitude of animal node using known position of the anchor node (B). The location along with the time stamp, sequence number and other information is stored in MYSQL database. These record are used by the system (PHP program running on base station) to reproduce the results on a web using Google map, displaying the position of the anchor nodes and that of the mobile nodes along with the unique sequence number as shown in figure 6.

Figure 7: RSSI and average RSSI vs. distance

As expected, from the experiments conducted, it was clear that the exact distance between the animal node and anchor node cannot be determined using RSSI values. However the errors can be minimized. Generally, these measurements are impacted by both time-varying and static environmentdependent errors. Time-varying errors (e.g., due to additive noise and interference) can be reduced by averaging multiple measurements over time. The variance found in the received signal strength has been found to be random but there

definitely seems to be a gradual decrease with increase in distance between the radio modules. The final system was tested on a small scale using four anchor nodes forming a cell and an animal node. The cell was of dimension 60m X 60m. Four anchor nodes (A, B, C, and D) were placed at the corners of the cell at a height of 6 feet. Animal node was kept 6 inches above the ground level. Prior to testing, RSSI value along each side of square at an interval of 3m was determined and stored in four 1 X 20 matrices. During the implementation of the system one of the anchor nodes was used as a base station. Dist(m)

3

6

9

12

15

18

21

24

27

30

AB

45

48

51

46

50

52

63

69

71

72

BC

41

45

48

54

56

58

60

63

64

66

CD

36

48

55

58

60

63

63

64

67

71

AD

48

53

52

44

49

52

5

57

64

64

Matrix

due to environmental conditions such as temperature, humidity and vegetation as follows: 1) In Free Space Region on 3rd of June, 2008 at 6 AM 2) Vegetation Region on 3rd of June, 2008 at 6 AM (The above 2 testings are performed simultaneously, the testing was done before the sunrise) 3) Vegetation Region: 4th of June, 2008 at 7:30 PM (This testing was done after the sunset) The results of the testing are depicted through the graphs as shown in figure 10 and figure 11. Through these testing’s we observed that the change in environmental condition has a significant effect on the RSSI values thus enhancing the error in tracking the location of the turtles. With the environmental deviations coming into play the four (1 X 20) calibration matrices will change accordingly. Therefore, an automation algorithm for dynamic calibration of environmental matrices has to be designed.

Figure 8.a Dist(m)

33

36

39

42

45

48

51

54

57

60

AB

61

72

65

67

66

67

67

79

79

78

BC

68

69

70

70

72

75

81

80

79

80

CD

73

75

63

65

64

66

65

66

70

71

AD

68

70

71

72

73

75

69

77

80

79

Matrix

Figure 8.b Figure 8: RSSI value vs. Distance with respect to four anchor nodes

As shown in figure 9, the animal node is moved following the dotted line path in the grid but the path actually tracked by the algorithm is different which is shown by solid line. In our experimental results, the mean error observed was 3.2m and 4.3m in x and y co-ordinate respectively. The peak error values were 6m in x co-ordinate and 8m in y coordinate.

Figure 10: RSSI values for free space and vegetation region

Figure 11: Variation of RSSI with time

Figure 9: System testing with four anchor nodes

The testing was done in the free space at DA-IICT football ground as shown in the figure 6. The system testing and calibration of the four sides was done keeping the same environmental conditions. Therefore to observe the effect of changing environmental conditions, a set of experiments were performed to study the variation in the signal strength (RSSI)

VIII. FUTURE WORK To abate the deviation due to changing environmental conditions an algorithm for automation of environmental matrix needs to be incorporated. One possible solution can be to scale the ideal calibration matrices based on change in environmental conditions. This can be done by inspecting the RSSI values of the packets send by the anchor nodes and comparing them with the corresponding values from the

matrices. This way a scaling factor can be calculated which can be used to adjust the matrix values. Also, nodes have to be optimized with respect to time, energy and space. REFERENCES [1] Prabhat Ranjan, Swetha Narumanchi, Pavan Kumar and Obulpathi Calla, “GPS less small turtle tracking system using sensor networks,”10th International Symposium on Wireless Personal Multimedia Communications, Jaipur, India, December 2007 [Proceeding]. [2] Chong Liu, Kui Wu, and Tian He; “Sensor localization with Ring Overlapping based on Comparison of Received Signal Strength Indicator”, 2004 IEEE International Conference on Mobile Ad-hoc and Sensor Systems (MASS), October 2004. [3] “ATMega32 Data Sheet,” www.atmel.com/dyn/resources/prod_documents/doc2503.pdf [4] “XBee manual,” http://www.maxstream.net/products/xbee/manual_xb_oem-rfmodules_802.15.4.pdf