Target Tracking a Non-Linear Target Path Using Kalman Predictive Algorithm by James Dennis Musick
Agenda Introduction Problem Definition Centroid Algorithm Kalman Filter Target Discrimination Conclusion Future Work
Introduction In the field of biomechanical research there is a subcategory that studies human movement or activity by video-based analysis Markers used –Optical –RF –Passive reflective –Etc… Video based motion analysis 2D Analysis 3D analysis Golf swing example
Problem Definition In order to track the following have to be accomplished –Centroid calculation –Prediction –Discrimination
Problem Definition cont. Trials used –Walking Trial –Jumping Trial –Waving Wand Trial –Increasing complexity
Centroid Algorithm Introduction Scanning scheme
Centroid Algorithm cont. 640 x 480 –~ pixels 8-bit Gray-scale Block diagram ThresholdX/Y address location Target Discrimination Buffer Logic control and centroid calculation Centroid Value Memory
Centroid Algorithm cont. Threshold
Centroid Algorithm cont. x/y addressing
Centroid Algorithm cont. Target Pixel Discrimination Buffer –x_sum, y_sum, LS_target, RS_target, Bot_target, target_pixel_num
Centroid Algorithm cont. Logic Control and Centroid Calculation
Centroid Algorithm cont. Centroid Memory Buffer –Once a target is completed (defined as no pixels within the search criteria at the row just below the target), then the centroid data is stored in a memory array until the data is read out at the end of the number of pictures that are being analyzed. –The array would be structured in the following manner if there were three targets in each of 5 pictures: Target_Centroid_Array = (xy,Target #, Picture #) => (1:2, 1:3, 1:5).
Centroid Algorithm cont. Examples
Centroid Algorithm cont. Performance and Limitations –Three targets simultaneous –Total number
Centroid Algorithm cont. Measurement Uncertainty Correct (3.5,4)Correct (3.5,3) Blue missing (3.5,4)Red missing (3.8,3.17) Red missing (3.64, 4.21)
Kalman Filter Introduction –State Space representation
Kalman Filter cont.
Kalman Filter cont
Target Models: –Noisy Acceleration model
Kalman Filter cont Target Models: –Noisy Jerk model
Kalman Filter cont Selection of update time: T = 1
Kalman Filter cont b
Operation of the Kalman Filter
Kalman Filter cont Operation of the Kalman Filter
Kalman Filter cont Operation of the Kalman Filter
Kalman Filter cont Operation of the Kalman Filter
Kalman Filter cont Operation of the Kalman Filter
Kalman Filter cont Operation of the Kalman Filter
Target Discrimination Introduction –Goal
Target Discrimination Example
Target Discrimination Example cont
Target Discrimination Operation of algorithm
Target Discrimination Operation of algorithm cont
Target Discrimination Operation of algorithm cont Jumping Trial
Target Discrimination Operation of algorithm cont
Target Discrimination Occluded targets
Conclusion Centroid algorithm Kalman filter –Model Discrimination
Future Work Hardware implementation 3D application Other biomechanical target discrimination (segmentation, etc.) Other tracking application (space, robotics, etc.)