Object detection using reflective surfaces
Tracking an object included in a reflective surface by detecting the reflective surface included in one or more images included in a plurality of images by determining a location of the reflective surface in pixel coordinates and tracking the location of the reflective surface in the plurality of images. Real world locations of the object can be determined based on attributes of the reflective surface including a geometric class, extrinsic properties that relate the reflective surface to an environment, intrinsic properties describing the reflective surface without referring to the environment and calibration properties of the sensor.
1. A system, comprising:
a computer that includes a processor and a memory, the memory including instructions executable by the processor to:
acquire a plurality of images from a sensor;
detect a reflective surface included in the images by determining a location of the reflective surface in pixel coordinates wherein a plurality of locations of the reflective surface is detected and the geometric class, the extrinsic properties, and the intrinsic properties of the reflective surface are determined by inputting the plurality of images into a neural network;
determine a plurality of real world object locations of an object included in the reflective surface based on attributes of the reflective surface including a geometric class, extrinsic properties that relate the reflective surface to an environment, intrinsic properties describing the reflective surface without referring to the environment and calibration properties of the sensor; and
determine a probability of contact based on a closest approach between the object and the sensor when the object and the sensor are predicted to pass without contact.
2. The system of claim 1 , the instructions including further instructions to determine an object trajectory based on tracking the plurality of real world locations of the object.
3. The system of claim 2 , the instructions including further instructions to determine the probability of contact between the object and the sensor based on the plurality of real world locations, the object trajectory, a location of the sensor and a trajectory for the sensor.
4. The system of claim 3 , wherein the probability of contact between the sensor and the object is determined by the trajectory for the sensor based on real world location, speed, and direction of the sensor, and determining the trajectory for the object based on real world location, speed, and direction of the object.
5. The system of claim 3 , wherein the sensor is included in a vehicle and the probability of contact between the object and the sensor includes a probability of contact between the object and the vehicle.
6. The system of claim 5 , the instructions including further instructions to operate the vehicle based on the probability of contact between the object and the vehicle.
7. The system of claim 1 , wherein the geometric class includes one or more of planar, convex, concave, and irregular, the extrinsic properties include location, size and orientation of the reflective surface, and the intrinsic properties include an optical distortion model.
8. The system of claim 1 , wherein the calibration properties of the sensor include a sensor location, a sensor orientation, a sensor focal distance, and a sensor scale.
9. The system of claim 1 , wherein the reflective surface is one or more of a mirror, a surface of a puddle, a window, and a metallic surface.
10. The system of claim 1 , wherein real world locations of the reflective surface are determined based on output from the neural network and the calibration properties of the sensor.
11. The system of claim 10 , wherein the real world locations of the reflective surface are tracked by inputting the real world locations of the reflective surface into a Kalman filter.
12. The system of claim 1 , the instructions including further instructions to determine a region of interest included in the reflective surface included in the plurality images that include the object.
13. A method, comprising:
acquiring a plurality of images from a sensor;
detecting a reflective surface included in the images by determining a location of the reflective surface in pixel coordinates wherein a plurality of locations of the reflective surface is detected and the geometric class, the extrinsic properties, and the intrinsic properties of the reflective surface are determined by inputting the plurality of images into a neural network;
determining a plurality of real world locations of an object included in the reflective surface based on attributes of the reflective surface including a geometric class, extrinsic properties that relate the reflective surface to an environment, intrinsic properties describing the reflective surface without referring to the environment; and calibration properties of the sensor; and
determining a probability of contact based on a closest approach between the object and the sensor when the object and the sensor are predicted to pass without contact.
14. The method of claim 13 , further comprising determining an object trajectory based on the plurality of real world locations of the object.
15. The method of claim 14 , further comprising determining the probability of contact between the object and the sensor based on the plurality of real world locations of an object, the object trajectory, a location of the sensor and a predicted trajectory for the sensor.
16. The method of claim 15 , wherein the probability of contact between the sensor and the object is determined by the trajectory for the sensor based on real world location, speed, and direction of the sensor, and determining the trajectory for the object based on real world location, speed, and direction of the object.
17. The method of claim 15 , wherein the sensor is included in a vehicle and the probability of contact between the object and the sensor includes a probability of contact between the object and the vehicle.
18. The method of claim 13 , wherein the geometric class includes one or more of planar, convex, concave, and irregular, the extrinsic properties include location, size and orientation of the reflective surface, and the intrinsic properties include an optical distortion model.
19. The method of claim 13 , wherein the calibration properties of the sensor include a sensor location, a sensor orientation, a sensor focal distance, and a sensor scale.
20. The method of claim 13 , wherein the reflective surface is one or more of a mirror, a surface of a puddle, a window, and a metallic surface.