Methods and systems for video compressive sensing for dynamic imaging
A compressive sensing system for dynamic video acquisition. The system includes a video signal interface including a compressive imager configured to acquire compressive sensed video frame data from an object, a video processing unit including a processor and memory. The video processing unit is configured to receive the compressive sensed video frame data from the video signal interface. The memory comprises computer readable instructions that when executed by the processor cause the processor to generate a motion estimate from the compressive sensed video frame data and generate dynamical video frame data from the motion estimate and the compressive sensed video frame data. The dynamical video frame data may be output.
1. A compressive sensing system for dynamic video acquisition, comprising:
a video signal interface comprising a compressive imager configured to acquire compressive sensed video frame data from an object, wherein the compressive sensed frame data is generated by sampling a signal associated with the object at rate below a Nyquist frequency of the signal;
a video processing unit comprising a processor and memory,
wherein the video processing unit is configured to receive the compressive sensed video frame data from the video signal interface,
wherein the memory comprises computer readable instructions that when executed by the processor cause the processor to:
generate a motion estimate from the compressive sensed video frame data;
generate dynamical video frame data from the motion estimate and the compressive sensed video frame data;
output the dynamical video frame data,
wherein the motion estimate is based on, at least in part, a velocity field,
wherein the velocity field is determined based on preview frames that are generated from a low frequency domain of a subset of the frames of the compressive sensed video frame data.
2. The compressive sensing system of claim 1 , wherein the motion estimate is obtained by solving an optimization problem comprising a compressible optical flow constraint.
3. The compressive sensing system of claim 2 , wherein the compressible optical flow constraint enforces the following condition:
∂ t u+q·∇u+u∇·q= 0
where u represents a dynamical video frame at time t, and q(x) represents the velocity field.
4. The compressive sensing system of claim 1 , wherein the compressive sensed video frame data comprises:
a first frame that specifies a scene at a first time, wherein the scene at the first time comprises the object at a first location within the scene; and
a second frame that specifies the scene at a second time, wherein the scene at the second time comprises the object at a second location within the scene,
wherein the first location and the second location are different locations within the scene.
5. A method for dynamic video acquisition, comprising:
obtaining frames of compressive sensed data for an object, wherein the compressive sensed data is generated by sampling a signal associated with the object at rate below a Nyquist frequency of the signal;
generating motion estimates using the frames of compressive sensed data; and
generating dynamical video frame data using the motion estimates and the frames of compressive sensed data by solving an optimization problem, the optimization problem comprising a wavelet transformation term and a total variation regularization term subject to a compressible optical flow constraint,
wherein the motion estimates are based on, at least in part, a velocity field,
wherein the velocity field is determined based on preview frames that are generated from a low frequency domain of a subset of the frames of the frames of compressive sensed data.
6. The method of claim 5 , wherein the compressible optical flow constraint requires solving the optimization problem such that:
∂ t u+q·∇u+u∇·q= 0
where u represents a decompressed dynamical video frame at time t, and q represents the velocity field that quantifies the motion estimate.
7. The method of claim 5 , further comprising:
generating preview frames from the frames of compressive sensed data using a multi-scale sensing matrix;
wherein the motion estimate is generated using the preview frames.
8. A method for dynamic video acquisition, comprising:
acquiring, by a compressive imager, frames of compressive sensed data, wherein the compressive sensed data is generated by sampling a signal associated with an object at rate below a Nyquist frequency of the signal;
generating motion estimates using the acquired frames of compressive sensed data; and
generating dynamical video frame data using the motion estimates and the frames of compressive sensed data by solving an optimization problem using an alternating direction augmented Lagrangian method,
wherein the motion estimates are based on, at least in part, a velocity field,
wherein the velocity field is determined based on preview frames that are generated from a low frequency domain of a subset of the frames of compressive sensed data.
9. A non-transitory computer readable medium comprising instructions that, when executed by a processor cause the processor to:
generate motion estimates using acquired frames of compressive sensed data, wherein the compressive sensed data is generated by sampling a signal associated with an object at rate below a Nyquist frequency of the signal; and
generate dynamical video frame data using the motion estimates and the frames of compressive sensed data by solving an optimization problem, the optimization problem comprising a wavelet transformation term and a total variation regularization term subject to a compressible optical flow constraint,
wherein the motion estimates are based on, at least in part, a velocity field,
wherein the velocity field is determined based on preview frames that are generated from a low frequency domain of a subset of the frames of compressive sensed data.
10. The non-transitory computer readable medium of claim 9 , further comprising instructions that, when executed by a processor cause the processor to:
generate preview frames from the frames of compressive sensed data using a multi-scale sensing matrix;
wherein the motion estimate is generated using the preview frames.
11. A non-transitory computer readable medium comprising instructions that, when executed by a processor cause the processor to:
generate motion estimates using acquired frames of compressive sensed data, wherein the compressive sensed data is generated by sampling a signal associated with an object at rate below a Nyquist frequency of the signal; and
generate dynamical video frame data using the motion estimates and the frames of compressive sensed data by solving an optimization problem using an alternating direction augmented Lagrangian method,
wherein the motion estimates are based on, at least in part, a velocity field,
wherein the velocity field is determined based on preview frames that are generated from a low frequency domain of a subset of the frames of compressive sensed data.