Proof of concept, Resolution, Future Advances in resolution, Speed, Calibration
SPEED
Speed
The proposed technology provides filtering without calculations. It sets upper and lower received intensities for each prime or IR and if all the primes are detected within the set values, it declares that a pixel is detected. This is done in digital hardware with no equation solving for detection of a target. Because of this, detection starts from the time that all the primary values are available to the filter plus 20 nanoseconds that is the propagation delays of the electronic components. Timing analysis of the proposed hardware based Tunable Filter (Figure 1) indicates filtering speeds of 80 milliseconds for a frame (or picture) of 4,000,000 pixels. This is a breakthrough processing times, to detect hundreds of targets compared to the present DSP practices.
Differences in Speed compared with present DSP Practices
The current concepts of FFT and Kalman filters extensively use multiplication and addition in either software or hardware to perform detection and thus long delays. As explained above their resolution is rather poor. Lack of resolutions in detection, most often forces the sensor platform to send raw data to a “center” for post processing and decision making. The delays due to the pre and post processing of data are a hindrance to real time immediate operations.
The results of studies of the present methodology with respect to speeds are outlined below:
The proposed technology provides filtering without calculations. It sets upper and lower received intensities for each prime or IR and if all the primes are detected within the set values, it declares that a pixel is detected. This is done in digital hardware with no equation solving for detection of a target. Because of this, detection starts from the time that all the primary values are available to the filter plus 20 nanoseconds that is the propagation delays of the electronic components. Timing analysis of the proposed hardware based Tunable Filter (Figure 1) indicates filtering speeds of 80 milliseconds for a frame (or picture) of 4,000,000 pixels. This is a breakthrough processing times, to detect hundreds of targets compared to the present DSP practices.
Differences in Speed compared with present DSP Practices
The current concepts of FFT and Kalman filters extensively use multiplication and addition in either software or hardware to perform detection and thus long delays. As explained above their resolution is rather poor. Lack of resolutions in detection, most often forces the sensor platform to send raw data to a “center” for post processing and decision making. The delays due to the pre and post processing of data are a hindrance to real time immediate operations.
The results of studies of the present methodology with respect to speeds are outlined below:
- Kalman filtering is probably the most commonly used algorithm for implementing the tracker, although recently Condensation algorithm [7] and mean shift algorithm [4] have shown to provide certain advantages especially in the presence of significant background clutter. The Kalman filter is primarily used for identification and tracking slow moving targets [11, 12, and 13]. Tracking is enhanced by the structural information perceived from the moving objects, which improves the classification [13]. Tracking is initiated every time a new moving object is determined in the scene and the features found to fulfill the predefined criteria [13].
- Kalman filter is considered to be too computationally infeasible for image super-resolution due to the size of the images and thus the state space involved [9]. It is assumed that a person changes its walking direction and walking speed according to Gaussian distributions, thereby, the translational velocity is assumed to lie between 0 and 150 cm/s.
- Variations of this filter have a widespread use in detections of moving cars in a parking lot, but they are not suited to track and count vehicles in a fast moving street or highway [12]. Tracking non-rigid targets in low-resolution images has long been realized as a region based correspondence problem, in which each target is mapped from one frame to the next according to its position, dimension, color and other contextual information. When multiple targets exist and their dimensions are not negligible in comparison with their velocities, occlusion or grouping of these targets is a routine event. This brings about uncertainty for the tracking, because the contextual information is only available for the group and individual targets cannot be identified. [10] [14].
- The proposed technology provides filtering without calculations. It sets upper and lower received intensities for each prime or IR and if all the primes are detected within the set values, it declares that a pixel is detected. This is done in digital hardware with no equation solving for detection of a target. Because of this, detection starts from the time that all the primary values are available to the filter plus 20 nanoseconds that is the propagation delays of the electronic components. Timing analysis of the proposed hardware based tunable filter (Figure 4) indicates filtering speeds of 80 milliseconds for a frame (or picture) of 4,000,000 pixels. This is a breakthrough processing times, to detect hundreds of targets compared to the present DSP practices. This feature should be one of the NASA’s objectives for robotic programs and discoveries of outer space targets.