System and method for detection and classification using a statistical feature vector
The present application discloses a method, system, and computer system for performing object detection and/or classification. The method includes (i) obtaining sensor data, (ii) processing the sensor data to obtain hypercube data representing the sensor data, (iii) transforming a data representation of the sensor data to obtain a statistical feature vector representing the sensor data, wherein the data representation is transformed based at least in part on processing the data representation using a Multistage Wiener Filter (MWF), (iv) performing an object detection and/or classification based at least in part on analyzing the statistical feature vector using a machine learning model, and (v) providing an indication of the object detection and/or classification.
1 . A system, comprising:
one or more processors configured to:
obtain sensor data;
transform the sensor data to a phase history domain representation;
sample the phase history domain representation using a predefined matrix;
generate a datacube based on output from the sampling of the phase history domain representation using the predefined matrix;
transform the datacube to obtain a statistical feature vector representing the sensor data, wherein the datacube is transformed based at least in part on processing the datacube using a Multistage Wiener Filter (MWF);
perform an object detection and/or classification based at least in part on analyzing the statistical feature vector using a machine learning model; and
provide an indication of the object detection and/or classification; and
a memory coupled to the one or more processors and configured to provide the one or more processors with instructions.
2 . The system of claim 1 , wherein the sensor data comprises synthetic aperture radar data.
3 . The system of claim 1 , wherein the processing the datacube using the MWF comprises determining one or more projections of the datacube based at least in part on one or more predefined steering vectors.
4 . The system of claim 1 , wherein the sampling the phase history domain representation using the predefined matrix includes to:
position the predefined matrix over the phase history domain representation; and
obtain a set of phase history values for the sensor data at locations at which the predefined matrix overlays the phase history domain representation.
5 . The system of claim 4 , wherein the sampling the phase history domain representation using the predefined matrix includes iteratively sampling regions of the sampled phase history domain representation by positioning the predefined matrix over a region of the phase history domain representation and obtaining associated phase history values based on the positioning of the predefined matrix until all regions have been sampled.
6 . The system of claim 1 , wherein the processing the datacube using the MWF comprises truncating a filtering based on the MWF to cause a size of expected resultant data to match a filter order of the MWF.
7 . The system of claim 1 , wherein the object detection and/or classification is provided contemporaneous with collection of the sensor data to provide tracking.
8 . The system of claim 1 , wherein performing the object classification includes comparing the statistical feature vector to object data comprised in a template library.
9 . The system of claim 8 , wherein the performing the object classification further includes detecting an object in the sensor data in response to determining that the statistical feature vector matches a sample of the object data comprised in the template library.
10 . The system of claim 9 , wherein the sensor data is associated with one or more objects.
11 . The system of claim 9 , wherein the sample of the object data corresponds to a statistical feature vector for a particular object.
12 . The system of claim 9 , wherein the template library comprises samples of object data for a plurality of different objects.
13 . The system of claim 9 , wherein the template library comprises samples of object data for a particular object.
14 . The system of claim 1 , wherein processing the datacube using the MWF comprises transforming resultant data output from the MWF into the statistical feature vector.
15 . The system of claim 1 , wherein the one or more processors are further configured to perform a rotation or scaling of the sensor data or the datacube before transforming the datacube.
16 . The system of claim 15 , wherein rotating or scaling the sensor data or the datacube is based on an alignment of samples of object data in a template library to which the statistical feature vector is compared.
17 . The system of claim 16 , wherein the alignment has a preferred axis.
18 . The system of claim 16 , wherein the scaling is based at least in part on a normalized scale size.
19 . The system of claim 1 , wherein:
raw data collected by one or more sensors or an image formed from the raw data is sliced into a set of data slices; and
the sensor data that is processed to generate the datacube corresponds to a single data slice from the set of data slices.
20 . The system of claim 19 , wherein:
performing the object classification includes comparing the statistical feature vector to object data comprised in a template library; and
a plurality of the set of data slices are subject to parallel processing to perform object classification based on respective statistical feature vectors that are obtained based on transforming respective datacube data corresponding to the plurality of the set of data slices.
21 . The system of claim 1 , wherein the indication of the object classification is used in connection with computer vision for autonomous navigation.
22 . The system of claim 1 , wherein values comprised in the statistical feature vector are characterized by values of weights, a set of predefined steering vectors, and null spaces for the set of predefined steering vectors.
23 . The system of claim 1 , wherein the one or more processors are further configured to:
train a model to perform the object classification.
24 . The system of claim 23 , wherein the model is a machine learning model.
25 . A method, comprising:
obtaining, by one or more processors, sensor data;
transforming the sensor data to a phase history domain representation;
sampling the phase history domain representation using a predefined matrix;
generating a datacube based on output from the sampling of the phase history domain representation using the predefined matrix;
transforming the datacube to obtain a statistical feature vector representing the sensor data, wherein the datacube is transformed based at least in part on processing the datacube using a Multistage Wiener Filter (MWF);
performing an object detection and/or classification based at least in part on analyzing the statistical feature vector using a machine learning model; and
providing an indication of the object detection and/or classification.
26 . A non-transitory computer readable medium comprising computer instructions for:
obtaining, by one or more processors, sensor data;
transforming the sensor data to a phase history domain representation;
sampling the phase history domain representation using a predefined matrix;
generating a datacube based on output from the sampling of the phase history domain representation using the predefined matrix;
transforming the datacube to obtain a statistical feature vector representing the sensor data, wherein the datacube is transformed based at least in part on processing the datacube using a Multistage Wiener Filter (MWF);
performing an detection and/or object classification based at least in part on analyzing the statistical feature vector using a machine learning model; and
providing an indication of the detection and/or object classification.