Autonomous vehicle environmental perception software architecture
A process for sensing a scene. The process includes receiving sensor data from a plurality of sensor modalities, where each sensor modality observes at least a portion of the scene containing at least one of the objects of interest and generates sensor data conveying information on the scene and of the object of interest. The process further includes processing the sensor data from each sensor modality to detect objects of interest and produce a plurality of primary detection results, each detection result being associated with a respective sensor modality. The process also includes fusing sensor data from a first sensor modality with sensor data from a second sensor modality to generate a fused 3D map of the scene, processing the fused 3D map to detect objects of interest and produce secondary detection results and performing object level fusion on the primary and the secondary detection results.
1 . A process for sensing a scene containing objects of interest, comprising:
a. receiving sensor data from a first sensor modality, wherein the sensor data conveys information on a portion of the scene and of at least one object of interest;
b. processing the sensor data derived from the first sensor modality to determine a category of the scene based on the sensor data derived from the first sensor modality;
c. selecting an operational mode based on the determined category;
d. configuring a second sensor modality to operate in the selected operational mode;
e. receiving sensor data from the second sensor modality;
f. fusing the sensor data derived from the first sensor modality with the sensor data derived from the second sensor modality to generate a fused three-dimensional (3D) map of the scene;
g. processing the fused 3D map to detect objects of interest in the scene.
2 . A process as defined in claim 1 , wherein the first sensor modality includes a camera to produce image data describing the portion of the scene.
3 . A process as defined in claim 2 , wherein the second sensor modality includes a lidar to produce lidar data describing the portion of the scene.
4 . A process as defined in claim 3 , wherein the image data conveys a plurality of pixels, wherein the lidar data conveys a plurality of distance measurements, the fusing including associating distance values from the lidar data to pixels in the image data.
5 . A process as defined in claim 4 , wherein the fusing including for first pixels in the image data for which no distance values are available in the lidar data estimating distance values at least in part on the basis of distance values in the lidar data available for second pixels in the image data that are visually similar to the first pixels.
6 . A process as defined in claim 5 , wherein the camera has a higher resolution than the lidar.
7 . A process as defined in claim 6 , wherein the fusing includes defining segments in the image data based at least in part on visual similarity of pixels in a segment.
8 . A process as defined in claim 7 , wherein the fusing defines boundaries of a given segment such as to encompass pixels depicting a portion of the scene with similar colors.
9 . A process as defined in claim 7 , wherein the fusing identifies three or more distance values in the lidar data associated with pixels in a particular segment and estimates on the basis of the plurality of distance values by using a plane equation, distance values for other pixels in the particular segment for which no distance values are available in the lidar data.
10 . A process as defined in claim 1 , wherein the second sensor modality includes a radar to produce radar data describing the portion of the scene.
11 . A process as defined in claim 1 , wherein the fusing further comprises generating an occupancy grid.
12 . A process as defined in claim 11 , wherein the occupancy grid defines objects of interest and relative positions of the objects of interest.
13 . A process as defined in claim 12 , wherein the objects of interest include vehicles.
14 . A process as defined in claim 13 , wherein the occupancy grid identifies a trajectory of individual vehicles.
15 . A process as defined in claim 14 , wherein the occupancy grid identifies a speed of individual vehicles.
16 . A process as defined in claim 15 , wherein the occupancy grid defines free space around the vehicles.
17 . A process as defined in claim 11 , wherein the occupancy grid identifies a motion path of detected objects in 3D space by tracking the detected objects through successive data frames.
18 . A process as defined in claim 1 , further comprising:
monitoring performance of the first sensor modality and performance of the second sensor modality in the scene respectively;
determining a first weight associated with the first sensor modality based on the monitored performance of the first sensor modality and a second weight associated with the second sensor modality based on the monitored performance of the second sensor modality; and
wherein the fusing comprises fusing the sensor data derived from the first sensor modality with the sensor data derived from the second sensor modality based on the first and second weight.
19 . A process as defined in claim 1 , further comprising:
determining whether the category of the scene changes;
in response to the category of the scene having changed, switching the selected operational mode to another operational mode from a library of operational modes.
20 . A system, comprising:
a. a first sensor operating according to a modality, the first sensor configured to observe at least a portion of a scene containing at least one of the objects of interest and to generate sensor data conveying information on the portion of the scene and of the at least one object of interest;
b. a perception functional block for processing the sensor data derived from the first sensor to determine a category of the scene based on the sensor data derived from the first sensor and select an operational mode based on the determined category;
c. a second sensor configured to operate in the selected operational mode and acquire sensor data during the selected operational mode;
d. a sensor fusion module configured for;
i. fusing the sensor data derived from the first sensor with the sensor data derived from the second sensor to generate a fused 3D map of the scene;
ii. processing the fused 3D map to detect objects of interest in the scene.
21 . A system as defined in claim 20 , wherein the first sensor includes a camera to produce image data describing the portion of the scene.
22 . A system as defined in claim 21 , wherein the second sensor includes a lidar to produce lidar data describing the portion of the scene.
23 . A system as defined in claim 22 , wherein the image data conveys a plurality of pixels, wherein the lidar data conveys a plurality of distance measurements, the sensor fusion module configured for associating distance values from the lidar data to pixels in the image data.
24 . A system as defined in claim 23 , wherein the sensor fusion module is configured for estimating for first pixels in the image data for which no distance values are available in the lidar data, distance values at least in part on the basis of distance values in the lidar data available for second pixels in the image data that are visually similar to the first pixels.
25 . A system as defined in claim 24 , wherein the camera has a higher resolution than the lidar.
26 . A system as defined in claim 25 , wherein the sensor fusion module is configured for defining segments in the image data based at least in part on visual similarity of pixels in a segment.
27 . A system as defined in claim 26 , wherein the sensor fusion module is configured to define boundaries of a given segment to encompass pixels depicting a portion of the scene with similar colors.
28 . A system as defined in claim 26 , wherein the sensor fusion module is configured for identifying three or more distance values in the lidar data associated with pixels in a particular segment and estimates on the basis of the plurality of distance values by using a plane equation, distance values for other pixels in the particular segment for which no distance values are available in the lidar data.
29 . A system as defined in claim 20 , wherein the sensor fusion module is configured for generating an occupancy grid.
30 . A system as defined in claim 29 , wherein the occupancy grid defines objects of interest and relative positions of the objects of interest.
31 . A system as defined in claim 30 , wherein the objects of interest include vehicles.
32 . A system as defined in claim 31 , wherein the occupancy grid identifies a trajectory of individual vehicles.
33 . A system as defined in claim 31 , wherein the occupancy grid identifies a speed of individual vehicles.
34 . A system as defined in claim 29 , wherein the occupancy grid identifies a motion path of detected objects in 3D space by tracking the detected objects through successive data frames.