Payload systems and tracking algorithms for photogrammetric measurement of parachute shape Mark R. Shortisa, Stuart Robsonb, Tom W. Jonesc, Benny Lunsfordc and João F. Oliveirad a
School of Mathematical and Geospatial Sciences, RMIT University, GPO Box 2476V, Melbourne 3001, Australia.
b
Department of Civil, Environmental and Geomatic Engineering, University College London, Gower Street, London WC1E 6BT, U.K. c
Advanced Sensing and Optical Measurement Branch, NASA Langley Research Center, Hampton VA 23681, U.S.A. d
INESC-ID, Rua Alves Redol 9, Apartado 13069, 1000-029 Lisboa, Portugal. ABSTRACT
Parachute systems play a critical role in many science and military missions. Currently, NASA and the U.S. Army air delivery systems programs are evaluating measurement technologies to support experimental and qualification testing of new and modified parachute concepts. Experiments to validate the concept of parachute shape measurement have been conducted in a controlled, indoor environment using both fixed and payload cameras. The paper will provide further detail on the rationale for the experiments, the design of the payload systems, the indoor and outdoor testing, and the subsequent data analysis to track and visualise the shape of the parachute. Keywords: Target tracking, image sequence, shape measurement, search algorithm, visualization 1 INTRODUCTION Aerodynamicists are developing new parachute concepts for replacement of the standard U.S. Army T10 canopy which dates back to the Korean War (see figure 1). The evolution of the modern soldier has seen the overall combined payload weight reach the limits of the older canopies. To validate air flow models for their new concepts, the Army has begun a series of both ground based and air drop tests of scalable parachute models for database development and validation of the mathematical modelling. The Army has sought external expertise in various measurement technologies to support this effort. The technology development activity is seen to have generic benefit where the entry, descent and landing systems incorporate a parachute. NASA is currently developing two new Crew Launch Vehicles, Ares and Orion, for missions such as manned explorations of Mars. These vehicles will rely on a parachute based landing similar to the system used for Apollo missions (see figure 1). Validation testing to measure the physical parameters involved during the inflation and decent phases of parachutes is a key requirement of the design of the landing system. Historically the shape of the canopy during inflation and descent has been determined qualitatively, however modern video imaging technology has provided a tool capable of supporting quantitative measurement. LaRC has a long history of optical systems and photogrammetric measurement of aerospace models, dating back to the 1970s1. Most recently this technology has been applied to the In-Space Propulsion programs 20m Solar Sail ground validation test program2. The two programs share many similarities in terms of metrology; their large scale and “membrane like” structure exhibit very non-linear aspects during deployment and inflation. A goal of this effort was the development of a scalable imaging system that supports ground based platform drops, wind tunnel testing and medium altitude air drops. The objectives of these measurements are to monitor the aerodynamics and structural dynamics of parachutes and to provide a database for the verification and validation of the mathematical models developed to predict the performance
[email protected]; phone +61 3 9925 9628; www.rmit.edu.au/mathsgeo
of these parachute concepts. Scale model ground testing of parachute systems has been conducted in both wind tunnels and low altitude air drops, however the efforts here are aimed at correlating shape information from air drops with the other parameters including acceleration, velocity, load and canopy configurations.
Figure 1a. Standard US Army T10 parachute canopy.
Figure 1b.
An example of new parachute design.
Figure 1c. Apollo 15 landing, similar in concept to the new NASA Crew Launch Vehicle.
2 INITIAL CONFIGURATION AND TESTING A program of testing was envisaged at the commencement of the project in order to incrementally develop the video capture systems and the processing techniques to track the shapes of parachutes. Whilst unconstrained parachute drops under uncontrolled conditions were, and still are, the ultimate aim of the research, the first step in the testing was to be in a controlled environment to allow repeatable, verifiable experiments. It was also accepted very early in the design process that the camera configuration required to record the video sequences would have to be a payload based system. With unconstrained, medium altitude parachute drops as the realistic testing scenario, cameras on the payload would be the most versatile and universal system. However the clear disadvantage of a payload system is that the frame of reference for the cameras is in motion and the tracking would be restricted to relative movement and shape changes only. The absolute path of the parachute through the air column can only be determined if there are external cameras, either on the ground or another airborne platform. Ground based cameras are problematic because of the very large changes in image scale for a parachute drop from realistic heights. The camera system would have to be computer controlled for zoom and orientation, which leads to a very complex calibration requirement. An airborne camera system would require a GPS/inertial platform to position and orient the camera system, as well as automated tracking control. Either of these systems are technically feasible but beyond the initial scope and budget for the project. A partial solution to the absolute position of the parachute during the drop is to include suitable sensors on the payload, such as an inertial system and air speed monitor. However the parachute and payload, in general, will not act as a completely rigid body and the orientation and position of the payload will therefore not fully reflect that of the parachute. Nevertheless this is a tractable solution that is feasible within the limits of the project.
2.1 Indoor Experiment Setup Given all of these considerations, initial tests were designed and carried out indoors as controlled experiments. A series of tests with a 1.8m drogue parachute (6 feet diameter) and a generic circular 2.3m parachute scale model (7.5 feet diameter) provided by the Army were conducted at NASA Langley Research Center (LaRC) and in the Plum Brook Space Power Facility (SPF) at NASA Glenn Research Center. The tests at LaRC were conducted as a trial to validate the approach used for image sequence capture, camera calibrations and the sequence processing. SPF has the capability to provide a controlled environment of pressure and temperature that may be used in the future to replicate high altitude conditions. The tests at SPF were conducted in ambient conditions, however the chamber was maintained at constant temperature and has the added benefit of minimisation of air currents to eliminate any influence on the parachute during test drops. SPF has the additional advantage of a longer drop time from the 40m ceiling, as compared to 25m at LaRC. Gantry and resection targets
Release mechanism
7.5 ft canopy Canopy on guide wire Guide wire
Drop point and floor cameras
Figure 2a. Experimental set up at Plum Brook Space Power Facility.
Payload
Figure 2b. Parachute scale model in the pre-drop position.
The drogue parachute comprised a conventional nylon fabric canopy made up of 12 pseudo-triangular sections known as gores, connected to the payload by nylon support cords. The scale model parachute is manufactured from continuous fabric, with 16 support cords. In order to increase visibility of the canopy, the cord support lines were replaced by thin steel wires. The payload, described in the next section, acted both as ballast and an instrumentation package. The parachute was dropped along a guide wire attached to the ceiling and tensioned using 100kg of weights suspended just above the floor. The guide wire passed unimpeded through the payload and the vent of the parachute. Figure 2 shows a schematic of the set up at SPF and the parachute scale model in the pre-drop position. The underside and leading edges, or skirts, of the canopies were covered with retro-reflective circular targets (see figure 3) which were illuminated by ring lights or near-axis lights on the cameras (see figure 4). Early test configurations adopted at LaRC for the drogue parachute delineated the gores and were spatially dense to ensure accurate determination of the canopy shape. However it became apparent that target tracking would be problematic, especially in the critical period during inflation of the parachute. Later configurations on the scale model parachute used less targets on the canopy or targets only on the skirt. Retro-reflective targets were also positioned near the connection point of the guide wire on the ceiling in each facility. The ceiling targets (see figure 6), including scale bars at SPF, were coordinated by a self-calibration network using a high resolution digital still camera and subsequently used as fixed references for the resection of the parachute images. The fixed image capture system comprised five Dalsa cameras on the floor surrounding the drop point. The Dalsa cameras have 1600 by 1200 resolution progressive scan sensors with 7.4 micrometre pixel spacing. The cameras have dual tap outputs which allow 30 frames per second read out rates via CameraLink interfaces. The cameras are synchronised based on a hardware trigger and the image sequences are captured using a multi-camera recording system from IO Industries P/L. The system provides a timestamp for synchronization checks in the corner of the recorded AVI image sequences. The synchronisation was validated at LaRC using a single retro-reflective target on a 100mm diameter laser beam “chopper” rotating at 200rpm. Based on simulations and the availability of 35mm and 50mm fixed lenses,
four cameras were fitted with 35mm lenses and one camera with a 50mm lenses. The camera with a 50mm lens was placed very close to the drop point to ensure continuous visibility of all canopy targets. The four cameras with 35mm lenses were placed in a circular pattern approximately 3m and 5m from the drop point for LaRC and SPF respectively (see figure 4). These locations were adopted as a compromise between the geometric strength of the photogrammetric network and optimisation of the field of view in terms of the extent of the parachute drop and occlusions caused by the curvature of the canopy near the leading edge. A sixth Dalsa camera was used at SPF but was unavailable for LaRC.
Figure 3. Typical target pattern used for most drop tests.
Figure 4. Overview of fixed cameras at the LaRC drop point.
2.2 Canister Payload System for Indoor Drops A purpose-built, self-contained canister has been developed as the payload for indoor testing to provide ballast and capture digital images. The package provides two compartments; one for instrumentation, and the other for the calibrated weights. The 150mm wide payload provides a modest 140mm high compartment for instrumentation. A similar 140mm compartment is incorporated into the design to accommodate the ballast. The adjustable ballast weights provide the capability of evaluating the performance of the parachutes over a range of descent velocities. 5 to 24 volt power for all components is provided by a lithium-ion battery regulated through a power distribution card.
Figure 5a. Payload canister overview showing ballast at bottom.
Figure 5b. Circuit cameras lighting.
board and
Figure 5c. Digital video recorders.
Two “circuit board” CCD cameras with 6mm lenses are arranged in an upward-looking stereo configuration and record monochrome 720 by 480 pixels resolution, interlaced images at 30 frames per second. The images are captured via RS170 video interface to miniature digital video recorders (DVRs). The DVRs utilise solid state flash memory cards and a fixed 4 to 1 JPEG compression to reduce the storage requirements. Approximately 4 minutes of video can be stored for each camera on 1Gb flash cards. The image sequence is triggered by a 6 channel radio frequency controller which transmits simultaneously to both DVRs, ensuring a synchronised start to the recording. The stereo pairs are subsequently synchronised based on an on-screen timer. For the payload lighting system a 24 volt high-intensity multi-LED array was designed (see figure 5). The white light LEDs incorporate a non-standard lens to project a 52º divergence cone to encompass the canopy in its fully deployed shape. The LEDs are independently controlled by radio signal. In addition to economizing the power drain on the battery, this feature allows the lighting on/off operation to serve as a synchronization check for the recorded video from the on-board cameras and link to the image sequences from the ground based cameras. The canister also incorporates a number of other sensors, such as air velocity from a hot wire anemometer, a Micro-Electro-Mechanical Systems (MEMS) tri-axial accelerometer and four load cells on the suspension lines between the canopy and the payload. The analog data from these sensors is acquired through a miniature 16 channel acquisition card. 2.3 Calibration of Multi-Camera System The two independent sub-systems of cameras are calibrated separately. Self-calibration and relative orientation of the fixed cameras is determined by lowering a calibration fixture down the guide wire used for the parachute drops (see figure 6). In an ideal case, the calibration fixture should travel the full extent of the drop, however in practice the number of locations can be limited by the logistics in the facility. Locations at the start, mid-point and bottom of the drop, with rotations of the fixture, are a practical solution that provides a 3D self-calibration network. The relative orientation, subsequently derived from the camera stations in the network, represents the full range of motion and should minimise any systematic errors that may ensue from a more limited sample of the drop extent. Despite the expected integrity of this approach, ceiling resection targets were used whenever available because the results of intersections on the canopy targets was typically more favourable. Further investigation is needed, however the narrow, elongated geometry of the networks is not optimal in either case.
Figure 6. Calibration fixture being lowered along the guide wire at LaRC. Note also the ceiling targets used for floor camera resections.
Figure 7. Full frame image from a payload camera showing interline motion displacement.
This procedure was performed at LaRC, but could not be competed in full at SPF. To strengthen the self-calibration at SPF, a number of additional static images of the calibration fixture from all individual Dalsa cameras were incorporated into the network. This strategy improved only the camera calibration aspect of the self-calibration.
A self-calibration and relative orientation of the payload cameras was determined by using the stereo-cameras to image a static calibration fixture using a multi-station, convergent network. In this case the calibration fixture was positioned at the same range as the canopy and then moved around within the field of view to optimise the camera calibration and relative orientation. A significant disadvantage of the interlaced images provided by the payload cameras is the motion displacement of target images, shown clearly in figure 7. Accordingly the payload cameras were calibrated in frame mode but were used for measurement and tracking in field mode. The relationship between calibrations in frame and field modes can be readily incorporated into the calibration parameters3. 3 SEQUENCE PROCESSING 3.1 Conventional Approach Target tracking commences with a simple setup procedure which is largely operator controlled. For convenience each set of acquired images is treated as a distinct epoch of data (see figure 8). To commence, either a previously established relative orientation of the cameras is used, or a minimum of four control targets on the ceiling are identified on all images in the first epoch. For the resection, a direct solution4 is then used to compute initial exterior orientation parameters for each camera. These initial values are then automatically refined by means of a robust least squares estimation resection. Following a successful resection, the computed exterior orientations of each camera are used to back-drive to the imaged locations of any additional control targets. A further resection is automatically computed to ensure that all available information has been used to compute each camera location and orientation. The next step is the identification and computation of the projected targets on the canopy surface. Two images from the first epoch are selected and all corresponding targets manually identified and measured to provide a spatially meaningful numbering sequence. Following an intersection to determine the three dimensional locations of these targets, their imaged locations in any remaining images in the set can be simply computed by the same back-driving procedure used to find the control targets. The back-driving procedure is made more robust to occlusions by searching for each computed target according to its predicted image dimension and depth away from the camera. Any targets omitted on completion of this procedure, such as those inseparably close together or partially occluded by parachute support lines, are identified manually. The tracking method is based on knowledge of the prior locations of each target in up to three preceding epochs. On the basis of this information a set of motion equations1 can be used to predict the location of the target at the next epoch. Since initial values for the camera exterior orientation are known, the approximate imaged location of the target can be found. For each new epoch the tracking process progresses according to five steps: a) b) c) d) e)
Back-driving to the computed locations of known control points. Re-computation of the camera location by least squares resection. Back-driving to the predicted locations of the tracked targets for each image in turn. Computation, by intersection and then network solution, of the locations of all targets. Estimates of the next location of each tracked target are then computed.
A photogrammetric network adjustment for each epoch, incorporating all measurement data, is optionally computed as part of step (d) once the resection/intersection procedure has been completed for each epoch. This step was found to be advantageous since control information was visible in all images and the ability to simultaneously determine optimal fit between longer range control points and closer target points on the parachute canopy takes all available information into account. This approach is in contrast to the use of the pre-determined relative orientation, which is based on a limited sample of control targets and calibration fixture locations on the guide wire. As noted previously, the geometry of the network is non-optimal and the multi-camera relative orientation risks extrapolation into an uncontrolled object volume. The tracking process is repeated for each successive image set. In this way a set of canopy surface points can be rapidly acquired for each epoch. The process is also flexible in terms of the number of images used in each set and indeed the
number of image sets. Each extra image set is simply treated as an additional epoch of data to be sequentially processed. In this particular example it is convenient to maintain the identification of each tracked target in all epochs, since this allows continuity of canopy motion to be assessed and modelled.
Figure 8. Object space schematic plus camera views for one epoch in an image sequence from SPF (note that this set has been acquired at a test electronic flash synchronisation point, giving a much more brightly illuminated canopy than is typical). 3.2 Search-based Approach The conventional approach to tracking is applicable for targets on the parachute canopy that are temporally consistent, but can fail particularly when targets on the extremities of the parachute flip in and out of view or move too rapidly for the camera frame rate and predictive tracking process to operate. For example, some parachute structures exhibit a harmonic motion or pulsation which opens and closes the leading edge. As an expansion ceases and becomes a contraction, the frame rate of the cameras is insufficient to capture this as a gradual change. The back-driving based on motion predication can fail with the predicted target image location being outside the new extent of the parachute. An automated search based approach for combined target image correspondence and tracking is currently undergoing refinement. The method determines the minimum crossing point in 3D space for each possible target image group correspondence and efficiently establishes an optimal set of correspondences with the aid of an optimised search tree. The capability of the method is dependent on distinguishing each individual target image in an image scanning process and on the fidelity of the parameters in the photogrammetric model that describe the long narrow lines of sight between each camera and the parachute canopy. Initial results on test objects are highly encouraging, but further development remains to be done in order to isolate occluded targets that are predominant at the parachute edges and to incorporate this information into the search tree. For example, a second pass may be performed. Target images that were not grouped with other observations in the first pass may be compared and assigned matches with the original data, even if the matches are weaker than the optimum target image grouping and identification sequence. If more targets can be computed with the addition of new correspondences
and deletion of previous correspondences, the identifications of the second pass are retained. The developed algorithm is a work in progress and will be reported in a future paper. 4 RESULTS AND VISUALISATION Results from a network adjustment incorporating ground cameras, compared to both ground and payload cameras, are shown in table 1. It is evident that the addition of the two payload cameras into the network confers a significant improvement in the attainable precision. This is unsurprising given the imaging geometry seen in figure 8, where the upper two cones represent the payload cameras and demonstrate their much closer proximity to the parachute. The relatively large coordinate precisions from an externally constrained network which utilises control targets to define the datum result from the fact that the control targets are located on the ceiling several metres above the parachute canopy. In contrast, optimal precisions from an inner constraints network indicate precisions of targets on the parachute of the order of 1 to 6mm. In the case of the free network solution the payload cameras degrade the plan precisions of the target coordinates because of a reduced internal consistency, largely caused by the poor image quality of the payload camera images. Within an individual epoch, the level of precision attainable utilising inner constraints enables a much finer resolution for shape analysis. Table 1. Comparisons of results for the various network solutions. External constraints network RMS image residual (micrometres) Number of redundancies in the network Relative precision for the network Mean precision of canopy targets (mm) RMS corrections on control (mm) Inner constraints network Mean precision of canopy targets (mm)
6 ground and 2 payload cameras 4.42 816 1 : 318 X Y Z 155 123 79 28 31 123 3.0
6.3
3.2
6 ground cameras 3.93 588 1 : 215 X Y Z 216 177 134 24 28 71 0.9
1.5
5.1
The particular parachute drops analysed in the tests at LaRC and SPF demonstrated a pulsation (“jelly fish”) or a twisting motion as they descended, creating additional difficulties for target tracking (see figure 9). In particular, targets at the periphery of the parachute edge moved closer together and then further apart as the canopy descended if pulsation was exhibited. A partial solution to the occlusion of one target by another (see figure 8) is simply to progressively reduce the dimensions of the search window used to find the targets in each image. In the case of occlusion, both targets are dropped from the tracking solution and then interactively re-acquired in a following epoch. This approach necessitated a tracking solution that required operator intervention in order to preserve consistent target numbering. Analysis of the observed motion typical of the parachutes under investigation is being undertaken in order to provide a better predictive model. An alternative, but currently less practical, solution is to increase the rate at which image sequences are acquired so that the relative image motion between data epochs is minimised. As noted in the previous section, a network solution was used in preference to the pre-determined relative orientation at each epoch. This option was validated by the testing with the relative orientation solution for the SPF sequences, which exhibited RMS image errors of the order of 10 to 15 micrometres, significantly larger than the 3 to 5 micrometer errors produced by the network adjustments. Whilst the use of the multi-camera relative orientations would simplify the sequence processing, the use of an externally constrained network solution is necessary in this case to provide a link between the ground and payload cameras, and provide a consistent datum for all epochs. A number of visualisations of the tracked targets have been generated to understand the changes in shape of the canopy, and in particular the variations of the leading edge or skirt of the parachute. The results presented here are extracted from the tracking carried out at SPF using the scale model parachute. Figure 9a shows the skirt with respect to a fixed coordinate datum as the parachute falls along the guide wire. The pulsations of the parachute soon after full inflation, easily identified in the image sequences, are evident. From this visualisation the magnitude of the pulsations can be estimated. Figure 9b shows the skirt with respect to the payload cameras. In this visualisation the lighter colours indicate progression in time. In some cases the leading edge of the canopy demonstrates a sinusoidal twisting motion relative to the payload as the parachute expands and collapses during the pulsations, and this effect was not readily
evident in the image sequences. In other cases the skirt exhibits a simple rotation. Further analysis of the shape changes and spatial behaviour of the canopy, along with the information from the other sensors on the payload, is required to fully understand the aerodynamics of the parachute.
Figure 9a. Side view of the skirt and canopy centre of a parachute with respect to a fixed datum – pulsation is clearly evident.
Figure 9b. Top view of the skirt and canopy centre of a parachute with respect to the payload cameras – sinusoidal and simple twisting.
5 OPEN AIR PARACHUTE DROPS 5.1 Air Drop Payload System The next stage of the test program moved from indoor testing with reduced scale parachutes to open air drops with a full size parachute. Accordingly, a much larger payload system could be designed to match the weight of a modern soldier and be robust enough to withstand the impact of landing on solid ground. The air drop payload is constructed of T-slotted, modular, aluminium framing manufactured by 80/20 Inc. The payload weighs approximately 80kg and has a height, width and length of 700 by 610 by 920mm. Additional ballast can be included to simulate heavier soldiers or added equipment. The payload was constructed to survive a ground impact at 6m/s. The frame must be supported with layers of 100mm cardboard honeycomb to prevent permanent frame
deformation as a result of ground impact. The payload has two pneumatically controlled arms for camera extension. The extendable arms provide the minimum base separation, with 10 degrees inward convergence, needed to support photogrammetric measurement of the parachute canopy. The payload camera arms are 480mm in length and provide an overall separation of 1.8m. The arms are extended for video recording once the risk of suspension line entanglement has receded and the arms are retracted before impact to prevent the risk of damage associated with payload roll-over. The air drop payload also includes several instrumentation systems intended to record the physical properties associated with the steady state phase of an air drop. The on-board instrumentation systems includes a hardened PC for overall control and data acquisition. In addition to the on-board cameras, the payload includes sensors for angle-of-attack, GPS and velocity. Data from these sensors is recorded using an 8-channel analog to digital signal recorder. The arm operation, lighting and data acquisition are remotely controlled using a ground-based radio control unit. The interline transfer circuit board cameras used with the first payload system have been updated to progressive scan Nexis cameras manufactured by Matrox. The 1/3” CCD sensors with 4.65 micrometre sensor spacing have a resolution of 1024 by 768 and a frame rate of up to 20Hz. The cameras use a custom digital interface to a Matrox 4Sight capture board which also manages the synchronisation of the cameras. The on-board PC is capable of recording one minute of image data at 20 frames per second from both cameras using a solid state drive. The payload cameras were outfitted with 4.8 mm lenses for the air drop tests. Two initial calibrations of the system were carried out. The first used a large indoor, 3D calibration range and the second used targets on the ceiling of a loading bay at approximately the expected range of the full size canopy. The calibrations produced consistent results with an RMS image residual of 0.1 pixels. A custom Hi-power LED ring light was incorporated in each camera package to provide the needed illumination (see figure 10). The LED ring lights operate in the visible red area of the spectrum (650nm) and to attenuate the impact of the ambient lighting each camera was fitted with a 630nm cut filter. These extra steps were necessary to reduce the impact of the ambient illumination based on the initial drop tests described in the next section.
Figure 10a. Schematic of the air drop payload.
Figure 10b. Air drop payload with impact protector and cameras stowed - detail of the camera inset.
5.2 Gantry Drop Tests The initial testing of the payload system was conducted using the Apollo moon lander test gantry at LaRC. The gantry is used routinely for a variety of impact tests from a height of up to 80m. The intent of the gantry drop tests was to ensure that the radio control, camera deployment, lighting and video capture systems operated effectively, and that the payload would withstand the impact with the ground, prior to an open air drop from a much higher altitude.
Eighty-four 50mm diameter retro-reflective targets were attached in a symmetrical pattern to the interior of the canopy of a standard U.S. Army 7m diameter parachute. Additionally, a 50mm target was folded over the leading edge of the canopy in the center of each gore to define the opening diameter during steady state conditions. The payload was fitted to the parachute and hauled up to the release point underneath the gantry (see figure 11). Three drops were conducted in the early morning hours in very still conditions. To ensure the safety of the drop, the payload was attached by cables to two vertical guide wires fixed between the gantry and the ground. All systems operated successfully and the payload maintained the structural integrity after each impact. The target images were adversely affected by ambient light through the semi-transparent canopy, and it is clear from the images (see figure 11) that remedial action would have to be taken to ensure the visibility of the targets.
Figure 11. Parachute at various stages during the moon lander gantry drops at LaRC, and an example image from the cameras. 5.3 Helicopter Drop Tests The next stage of the testing of the payload system has been low altitude air drops, conducted at the U.S. Army National Guard Training Center at Quonset Point, Rhode Island. Three drops were conducted from a UH-1 Huey helicopter at an altitude of 200m. The payload was held and released from a side mounted man-lift arm on the helicopter. To assist with the contrast of each target, a flexible backing material was developed to attach directly to the canopy and provide minimum stiffness so as to not affect the parachute performance significantly. The backing material was applied to the outside of the canopy behind each target to block out the ambient light around the target (see figure 12). The combination of the backing material and the red light illumination and cut filter largely eliminates the image problems
experienced with the gantry drops. The air drop testing has demonstrated that the payload continues to withstand the impact with the ground and the images are of high quality with no motion artefacts (see figure 12).
Figure 12a. Parachute released from the helicopter.
Figure 12b. Backing used for the retro-reflective targets.
Figure 12c. Example image from the payload cameras.
6 CONCLUSIONS The approach to the characterisation of the parachute models described in this paper is successful in extracting sequences of target coordinates from the imagery captured in the indoor, controlled experiments, but is currently limited by the resolution of the sensors and the quality of the images. In addition, the tracking solution requires significant manual intervention to process the sequences. The results for the image sequences captured for the open air drops has been improved significantly by the higher resolution of the sensors and higher quality images from the cameras. Further, the search based tracking of target images will provide a more effective process for the analysis image sequences with less manual intervention. Continuing progress in this project will be reported in future papers. Acknowledgements The research and development described in this paper is supported by the U.S. Army Natick Soldier Center and NASA Langley Research Center. This paper is partly based on a paper published in the proceedings of the Eighth Conference on Optical 3-D Measurement Techniques5. REFERENCES 1. 2.
3. 4. 5.
Shortis, M. R. and Snow, W. L., 1997. Videometric tracking of wind tunnel aerospace models at NASA Langley Research Centre. The Photogrammetric Record, 15(85): 673-689. Shortis M. R., Robson, S., Pappa, R. S., Jones, T. W. and Goad, W. K., 2002. Characterisation and tracking of membrane surfaces at NASA Langley Research Center. International Archives of Photogrammetry and Remote Sensing, 34(5): 90-94. ISSN 1682-1777. Shortis, M. R. and Snow, W. L., 1995. Calibration of CCD cameras for field and frame capture modes. Proceedings, Conference on Digital Photogrammetry and Remote Sensing '95, SPIE Vol. 2646, pp. 2-14. Zeng, Z. and Wang, X., 1992. A general solution of a closed-form space resection. Photogrammetric Engineering and Remote Sensing, 58(3): 327-338. Shortis, M. R., Robson, S., Jones, T. W. and Lunsford, C. B., 2007. Parachute model validation using image sequences. Eighth Conference on Optical 3-D Measurement Techniques, A. Grun and H. Kahmen (Eds), ETH Zurich, Switzerland, ISBN 3-906467-67-8, Volume I, pp 72-79.
