Presentation on theme: "Hi_Lite Scott Fukuda Chad Kawakami. Background ► The DARPA Grand Challenge ► The Defense Advance Research Project Agency (DARPA) established a contest."— Presentation transcript:
Hi_Lite Scott Fukuda Chad Kawakami
Background ► The DARPA Grand Challenge ► The Defense Advance Research Project Agency (DARPA) established a contest in 2004 to set a new level of accomplishment for autonomous land vehicles. This was in response to a goal set by Congress that by 2015 twenty percent of U.S. military vehicles on a battlefield would be unmanned. In The Grand Challenge, contestants vied for a one million dollar prize by racing for 150 miles.
HI_Lite Vehicle ► Modified from a Columbia Par Car electric scout vehicle ► Akamai Research LLC hosted and led the Hawaii effort with volunteers and sponsorship from e-Vehicles of Hawaii. ► In only three months the car was born. ► HI_Lite stayed in consideration up to the semi finals of the DARPA challenge.
Current status ► Vehicle is currently dismantled. ► No longer bounded by the DARPA contest rules. ► Looking to improve operational capabilities/ strategies and robotic controls, integrate wireless feedback and multiple remote controllers, including human operators, and allow for “urban” environments.
Our Project ► To explore the capabilities of the vehicle’s “eyes” and derive a primitive useful “environment describing” system. ► Coupling simple optical devices to create an enhanced “environment describing” system.
Block Diagram Software Small Vision System Matlab (or C++) Hardware Videre Design digital stereo Camera head Laser Pointers Lenses
Block Diagram - Hardware ► Videre Design Variable baseline digital stereo camera head Provides left and right video/images for depth/3D information ► Laser Pointers Create a 3D laser grid and illuminate grid on objects within the camera's view to calibrate depth information ► If necessary, external "pre-processing" will be done using color lenses as filters to enable the camera to see the lasers as lines instead of points ► If necessary, use the SICK Laser Range Detector to establish numerical calibration points and calibrate/tune the stereo camera using the SICK's data
Block Diagram - Software ► Small Vision System – Videre bundled software Record stereo images Rectify the images to account for distortion Perform stereo correlation to compute a range image (depth contours) ► Matlab - interpret data and provide useful outputs Object detection calculations using edge detection and boundary tracing on disparity images Correlate objects with numerical distance points from laser grid
Software – In Depth ► Object Detection ► Philsophy Simpler = Better! Why? ► We don’t need to record or identify objects we “see”, we only need to avoid hitting objects
Software – In Depth ► Object Detection ► Algorithm SVS library performs disparity calculations which give us depth information that can be converted to a 3D image
Software – In Depth ► Using the disparity image, look for edges that will distinguish object and trace the boundaries around each object
Software – In Depth ► To ensure a complete object, look for cohesion in objects depth by observing correlation between the pixels in the disparity image since the disparity image uses the brightness of the pixels to show depth (brighter = closer)
Software – In Depth ► Separate objects, then output object information and distance (determined from laser grid)
Software – In Depth ► Potential problems with algorithm: Tiny objects Overlaying objects Oddly shaped objects
Test Bed 5 Feet 5 1/2 Feet 6 Feet
5 Feet 5 1/2 Feet 6 Feet Test Bed 2 Filters Laser “array” distance
5 Feet 5 1/2 Feet 6 Feet Test Bed 3 Filters Laser “array”
5 Feet 5 1/2 Feet 6 Feet Test Bed 4 Filters Laser “array”
Work to be done ► Implement useful simple algorithm for object detection ► Test effects of coupling devices ► Move from image to video functionality ► Testing, testing, testing!
Hi-Lite Gant ChartScott Fukuda and Chad Kawakami 18-Feb25-Feb4-Mar11-Mar18-Mar25-Mar1-Apr8-Apr15-Apr22-Apr29-Apr Tests Setup Test Bed 1 Run Test Bed 1 Setup Test Bed 2 Run Test Bed 2 Data organize algorithm Exploratory additional Tests