System and method of perception error evaluation and correction by solving optimization problems under the probabilistic signal temporal logic based constraints
Described is a system to evaluate and reduce perception error in object detection and recognition. The system includes a perception module that receives perception data (of an object(s)) from an environment proximate a mobile platform. Perception probes are generated that describe one or more characteristics of the objects. The perception probes are converted into probabilistic signal temporal logic (PSTL)-based constraints that provide axioms having statistical analysis of the perception probes. The axioms are evaluated to classify the perception probes as valid or erroneous. Optimal perception parameters are generated by solving an optimization problem based on the axioms, which allows the system to adjust the perception module based on the optimal perception parameters.
1. A system to evaluate and reduce perception error in object detection and recognition, the system comprising:
one or more processors and a memory, the memory being a non-transitory computer-readable medium having executable instructions encoded thereon, such that upon execution of the instructions, the one or more processors perform operations of:
receiving, with a perception module, perception data from an environment, the perception data reflecting one or more objects in the environment;
generating perception probes from the perception data, the perception probes describing one or more characteristics of the one or more objects;
converting the perception probes into probabilistic signal temporal logic (PSTL)-based constraints, the PSTL-based constraints providing axioms having statistical analysis of one or more perception probes;
evaluating the axioms to classify the perception probes as valid or erroneous, wherein in evaluating the axioms to classify the perception probes as valid or erroneous, the axioms provide statistical separation between true positive and false positive distributions;
generating optimal perception parameters by solving an optimization problem based on the axioms; and
adjusting the perception module based on the optimal perception parameters.
2. The system as set forth in claim 1 , wherein adjusting the perception module includes causing a camera to initiate a physical action selected from a group consisting of changing a directional focus or altering a field-of-view.
3. The system as set forth in claim 1 , wherein generating perception probes from the perception data includes detecting the one or more objects in the perception data, tracking the one or more objects, and recognizing an action of the one or more objects.
4. The system as set forth in claim 1 , wherein in converting the perception probes into probabilistic signal temporal logic (PSTL)-based constraints, a probabilistic signal temporal logic for a probe f(x) is generated as follows:
∀ x,Pr ( a≤f ( x,t s :t e )≤ b→y )≥ P TP ,
where Pr(•) is a predicate and y is a true detection or recognition, where P TP denotes confidence probability of the probe, and where t s :t e denotes a time sequence between t s and t e , such that f(x, t s :t e ) is a probe sequence in time frame of t s :t e , and where a and b denote lower bound and the upper bound, respectively, of f( ) for statistical true positive distributions, and ∀x means for every detection/recognition x.
5. The system as set forth in claim 4 , wherein the perception probes are classified as erroneous when the confidence probability is outside of the lower bound or upper bound.
6. The system as set forth in claim 1 , wherein generating optimal perception parameters is determined by identifying an optimal control input u OPT that minimizes perception errors.
7. A computer program product for evaluating and reducing perception error in object detection and recognition, the computer program product comprising:
a non-transitory computer-readable medium having executable instructions encoded thereon, such that upon execution of the instructions by one or more processors, the one or more processors perform operations of:
receiving, with a perception module, perception data from an environment, the perception data reflecting one or more objects in the environment;
generating perception probes from the perception data, the perception probes describing one or more characteristics of the one or more objects;
converting the perception probes into probabilistic signal temporal logic (PSTL)-based constraints, the PSTL-based constraints providing axioms having statistical analysis of one or more perception probes;
evaluating the axioms to classify the perception probes as valid or erroneous, wherein in evaluating the axioms to classify the perception probes as valid or erroneous, the axioms provide statistical separation between true positive and false positive distributions;
generating optimal perception parameters by solving an optimization problem based on the axioms; and
adjusting the perception module based on the optimal perception parameters.
8. The computer program product as set forth in claim 7 , wherein adjusting the perception module includes causing a camera to initiate a physical action selected from a group consisting of changing a directional focus or altering a field-of-view.
9. The computer program product as set forth in claim 7 , wherein generating perception probes from the perception data includes detecting the one or more objects in the perception data, tracking the one or more objects, and recognizing an action of the one or more objects.
10. The computer program product as set forth in claim 7 , wherein in converting the perception probes into probabilistic signal temporal logic (PTSTL)-based constraints, a probabilistic signal temporal logic for a probe f(x) is generated as follows:
∀ x,Pr ( a≤f ( x,t s :t e )≤ b→y )≥ P TP ,
where Pr(•) is a predicate and y is a true detection or recognition, where P TP denotes confidence probability of the probe, and where t s :t e denotes a time sequence between t s and t e , such that f(x, t s :t e ) is a probe sequence in time frame of t s :t e , and where a and b denote lower bound and the upper bound, respectively, of f( ) for statistical true positive distributions, and ∀x means for every detection/recognition x.
11. The computer program product as set forth in claim 10 , wherein the perception probes are classified as erroneous when the confidence probability is outside of the lower bound or upper bound.
12. The computer program product as set forth in claim 7 , wherein generating optimal perception parameters is determined by identifying an optimal control input u OPT that minimizes perception errors.
13. A computer implemented method for evaluating and reducing perception error in object detection and recognition the method comprising an act of:
causing one or more processers to execute instructions encoded on a non-transitory computer-readable medium, such that upon execution, the one or more processors perform operations of:
receiving, with a perception module, perception data from an environment, the perception data reflecting one or more objects in the environment;
generating perception probes from the perception data, the perception probes describing one or more characteristics of the one or more objects;
converting the perception probes into probabilistic signal temporal logic (PSTL)-based constraints, the PSTL-based constraints providing axioms having statistical analysis of one or more perception probes;
evaluating the axioms to classify the perception probes as valid or erroneous, wherein in evaluating the axioms to classify the perception probes as valid or erroneous, the axioms provide statistical separation between true positive and false positive distributions;
generating optimal perception parameters by solving an optimization problem based on the axioms; and
adjusting the perception module based on the optimal perception parameters.
14. The method as set forth in claim 13 , wherein adjusting the perception module includes causing a camera to initiate a physical action selected from a group consisting of changing a directional focus or altering a field-of-view.
15. The method as set forth in claim 13 , wherein generating perception probes from the perception data includes detecting the one or more objects in the perception data, tracking the one or more objects, and recognizing an action of the one or more objects.
16. The method as set forth in claim 13 , wherein in converting the perception probes into probabilistic signal temporal logic (PSTL)-based constraints, a probabilistic signal temporal logic for a probe f(x) is generated as follows:
∀ x,Pr ( a≤f ( x,t s :t e )≤ b→y )≥ P TP ,
where Pr(•) is a predicate and y is a true detection or recognition, where P TP denotes confidence probability of the probe, and where t s :t e denotes a time sequence between t s and t e , such that f(x, t s :t e ) is a probe sequence in time frame of t s :t e , and where a and b denote lower bound and the upper bound, respectively, of f( ) for statistical true positive distributions, and ∀x means for every detection/recognition x.
17. The method as set forth in claim 16 , wherein the perception probes are classified as erroneous when the confidence probability is outside of the lower bound or upper bound.
18. The method as set forth in claim 13 , wherein generating optimal perception parameters is determined by identifying an optimal control input u OPT that minimizes perception errors.