Method for efficient target detection from images robust to occlusion
First Claim
1. A method for detecting targets from images which are captured by one or more cameras, said method comprising the steps of:
- a) collecting information about geometric and calibration parameters corresponding to each of the one or more cameras;
b) choosing a range of numerical representations defining a range of target configurations to be recognized from the images captured by the one or more cameras, wherein each instance in said range of numerical representations defines a corresponding target state of a plurality of target states associated with the range of numerical representations;
c) selecting a target model that is associated with the chosen range of numerical representations;
d) implementing an image projection procedure that overlays the target model onto an image to define a projected target model for each of the plurality of target states associated with the chosen range of numerical representations;
e) selecting a predetermined range of image features for each of the one or more cameras;
f) implementing a likelihood function between the projected target model and the predetermined range of image features;
g) selecting a first set of a predetermined number of the target states associated with the chosen range of numerical representations to be detected;
h) generating, for a first of the one or more cameras, a feature support map for each image feature in the predetermined range of image features;
i) projecting a first target state of the predetermined number of target states using the image projection procedure and associated calibration information from the first of the one or more cameras to form an image projection;
j) determining from the predetermined range of image features, a range of activated image features associated with a first activated pixel from the image projection procedure;
k) determining a first value associated with how the image feature is projected onto the image;
l) storing said determined first value paired with the projected target state of step (i) in each of the feature support maps associated with said range of activated image features;
m) repeating steps (j) through (l) for a next activated pixel that is activated by the image projection procedure;
n) repeating steps (i) through (m) for a next target state of the predetermined number of the target states to complete the feature support map of each image feature in the predetermined range of image features;
o) repeating steps (i) through (n) for a next camera of the one or more cameras; and
p) processing a first plurality of images captured by the one or more cameras to determine a probabilistic occupancy map from the feature support maps.
1 Assignment
0 Petitions
Accused Products
Abstract
The method for efficient target detection from images robust to occlusion disclosed by the present invention detects the presence and spatial location of a number of objects in images. It consists in (i) an off-line method to compile an intermediate representation of detection probability maps that are then used by (ii) an on-line method to construct a detection probability map suitable for detecting and localizing objects in a set of input images efficiently. The method explicitly handles occlusions among the objects to be detected and localized, and objects whose shape and configuration is provided externally, for example from an object tracker. The method according to the present invention can be applied to a variety of objects and applications by customizing the method'"'"'s input functions, namely the object representation, the geometric object model, its image projection method, and the feature matching function.
17 Citations
20 Claims
-
1. A method for detecting targets from images which are captured by one or more cameras, said method comprising the steps of:
-
a) collecting information about geometric and calibration parameters corresponding to each of the one or more cameras; b) choosing a range of numerical representations defining a range of target configurations to be recognized from the images captured by the one or more cameras, wherein each instance in said range of numerical representations defines a corresponding target state of a plurality of target states associated with the range of numerical representations; c) selecting a target model that is associated with the chosen range of numerical representations; d) implementing an image projection procedure that overlays the target model onto an image to define a projected target model for each of the plurality of target states associated with the chosen range of numerical representations; e) selecting a predetermined range of image features for each of the one or more cameras; f) implementing a likelihood function between the projected target model and the predetermined range of image features; g) selecting a first set of a predetermined number of the target states associated with the chosen range of numerical representations to be detected; h) generating, for a first of the one or more cameras, a feature support map for each image feature in the predetermined range of image features; i) projecting a first target state of the predetermined number of target states using the image projection procedure and associated calibration information from the first of the one or more cameras to form an image projection; j) determining from the predetermined range of image features, a range of activated image features associated with a first activated pixel from the image projection procedure; k) determining a first value associated with how the image feature is projected onto the image; l) storing said determined first value paired with the projected target state of step (i) in each of the feature support maps associated with said range of activated image features; m) repeating steps (j) through (l) for a next activated pixel that is activated by the image projection procedure; n) repeating steps (i) through (m) for a next target state of the predetermined number of the target states to complete the feature support map of each image feature in the predetermined range of image features; o) repeating steps (i) through (n) for a next camera of the one or more cameras; and p) processing a first plurality of images captured by the one or more cameras to determine a probabilistic occupancy map from the feature support maps. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17)
-
-
18. A computer program product for detecting targets from images which are captured by one or more cameras, the computer program product comprising a non-transitory computer readable medium having computer readable program code embodied therein that, when executed by a processor, causes the processor to:
-
a) collect information about geometric and calibration parameters corresponding to each of the one or more cameras; b) choose a range of numerical representations defining a range of target configurations to be recognized from the images captured by the one or more cameras, wherein each instance in said range of numerical representations defines a corresponding target state of a plurality of target states associated with the range of numerical representations; c) select a target model that is associated with the chosen range of numerical representations; d) implement an image projection procedure that overlays the target model onto an image to define a projected target model for each of the plurality of target states associated with the chosen range of numerical representations; e) select a predetermined range of image features for each of the one or more cameras; f) implement a likelihood function between the projected target model and the predetermined range of image features; g) select a first set of a predetermined number of the target states associated with the chosen range of numerical representations to be detected; h) generate, for a first of the one or more cameras, a feature support map for each image feature in the predetermined range of image features; i) project a first target state of the predetermined number of target states using the image projection procedure and associated calibration information from the first of the one or more cameras to form an image projection; j) determine from the predetermined range of image features, a range of activated image features associated with a first activated pixel from the image projection procedure; k) determine a first value associated with how the image feature is projected onto the image; l) store said determined first value paired with the projected target state of step (i) in each of the feature support maps associated with said range of activated image features; m) repeat steps (j) through (l) for a next activated pixel that is activated by the image projection procedure; n) repeat steps (i) through (m) for a next target state of the predetermined number of the target states to complete the feature support map of each image feature in the predetermined range of image features; o) repeat steps (i) through (n) for a next camera of the one or more cameras; and p) process a first plurality of images captured by the one or more cameras to determine a probabilistic occupancy map from the feature support maps. - View Dependent Claims (19, 20)
-
Specification