Free space detection for parking and driving in puddle areas with permuted fusion network
A method includes receiving one or more images, wherein at least one of the one or more images depicts a water region and analyzing, by one or more processors, the one or more images using a first machine learning model to determine a depth of the water region. The method also includes analyzing, by the one or more processors, the one or more images using a second machine learning model to determine a surface normal of the water region and performing, by the one or more processors, using a third machine learning model, multi-class segmentation of the one or more images. Additionally, the method includes performing one or more fusion operations on outputs of at least two of the first machine learning model, the second machine learning model and the third machine learning model to generate a classification of the water region.
1 . A method comprising:
receiving one or more images, wherein at least one of the one or more images depicts a water region;
analyzing, by one or more processors, the one or more images using a first machine learning model to determine a depth of the water region;
analyzing, by the one or more processors, the one or more images using a second machine learning model to determine a surface normal of the water region;
performing, by the one or more processors, using a third machine learning model, multi-class segmentation of the one or more images; and
performing one or more fusion operations on outputs of at least two of the first machine learning model, the second machine learning model and the third machine learning model to generate a classification of the water region.
2 . The method of claim 1 , further comprising: automatically determining, using a pre-defined set of rules, whether to park in a space containing the water region based on the classification of the water region and based on one or more parameters of a vehicle.
3 . The method of claim 1 , further comprising detecting one or more mirages.
4 . The method of claim 3 , wherein detecting the one or more mirages further comprises performing a fusion operation on outputs of the first machine learning model and the third machine learning model.
5 . The method of claim 1 , wherein performing multi-class segmentation further comprises classifying pixels of the one or more images into one or more pre-defined categories.
6 . The method of claim 5 , wherein the one or more pre-defined categories include at least hazardous pixels and non-hazardous pixels.
7 . The method of claim 1 , wherein the first machine learning model is trained using a training set of images and a plurality of depth maps corresponding to each image in the training set of images.
8 . The method of claim 1 , further comprising:
classifying the water region using a fourth machine learning model, wherein the classifying includes decoding segmentation of the outputs of the one or more fusion operations.
9 . The method of claim 7 , wherein the first machine learning model is trained to minimize an error between a predicted depth map and a ground truth depth map using an absolute error loss function and/or squared error loss function.
10 . The method of claim 9 , wherein the second machine learning model is trained to minimize a difference between angles of a predicted surface normal vector and a ground truth surface normal vector by using a cosine similarity loss function.
11 . The method of claim 7 , wherein performing the one or more fusion operations further comprises applying attention mechanism to keys and values.
12 . The method of claim 7 , further comprising: determining one or more uncertainty estimates for the third machine learning model using at least one of Monte-Carlo dropout and Laplacian techniques.
13 . An apparatus for detection and classification of water regions, the apparatus comprising:
a memory for storing one or more images; and
processing circuitry in communication with the memory, wherein the processing circuitry is configured to:
receive the one or more images, wherein at least one of the one or more images depicts a water region;
analyze the one or more images using a first machine learning model to determine a depth of the water region;
analyze the one or more images using a second machine learning model to determine a surface normal of the water region;
perform using a third machine learning model, multi-class segmentation of the one or more images; and
perform one or more fusion operations on outputs of at least two of the first machine learning model, the second machine learning model and the third machine learning model to generate a classification of the water region.
14 . The apparatus of claim 13 , wherein the processing circuitry is further configured to:
automatically determine, using a pre-defined set of rules, whether to park in a space containing the water region based on the classification of the water region and based on one or more parameters of a vehicle.
15 . The apparatus of claim 13 , wherein the processing circuitry is further configured to detect one or more mirages.
16 . The apparatus of claim 15 , wherein the processing circuitry configured to detect the one or more mirages is further configured to perform a fusion operation on outputs of the first machine learning model and the third machine learning model.
17 . The apparatus of claim 13 , wherein the processing circuitry configured to perform multi-class segmentation is further configured to classify pixels of the one or more images into one or more pre-defined categories.
18 . The apparatus of claim 17 , wherein the one or more pre-defined categories include at least hazardous pixels and non-hazardous pixels.
19 . The apparatus of claim 13 , wherein the first machine learning model is trained using a training set of images and a plurality of depth maps corresponding to each image in the training set of images.
20 . The apparatus of claim 13 , wherein the processing circuitry is further configured to classify the water region using a fourth machine learning model, wherein the classifying includes decoding segmentation of the outputs of the one or more fusion operations.
21 . The apparatus of claim 19 , wherein the first machine learning model is trained to minimize an error between a predicted depth map and a ground truth depth map using an absolute error loss function and/or squared error loss function.
22 . The apparatus of claim 21 , wherein the second machine learning model is trained to minimize a difference between angles of a predicted surface normal vector and a ground truth surface normal vector by using a cosine similarity loss function.
23 . The apparatus of claim 19 , wherein the processing circuitry configured to perform the one or more fusion operations is further configures to apply attention mechanism to keys and values.
24 . The apparatus of claim 19 , wherein the processing circuitry is further configured to determine one or more uncertainty estimates for the third machine learning model using at least one of Monte-Carlo dropout and Laplacian techniques.
25 . A non-transitory computer-readable medium storing instructions that, when applied by processing circuitry, causes the processing circuitry to:
receive one or more images, wherein at least one of the one or more images depicts a water region;
analyze the one or more images using a first machine learning model to determine a depth of the water region;
analyze the one or more images using a second machine learning model to determine a surface normal of the water region;
perform using a third machine learning model, multi-class segmentation of the one or more images; and
perform one or more fusion operations on outputs of at least two of the first machine learning model, the second machine learning model and the third machine learning model to generate a classification of the water region.
26 . A device for detection and classification of water regions, the device comprising one or more means for performing steps of:
receiving one or more images, wherein at least one of the one or more images depicts a water region;
analyzing, by one or more processors, the one or more images using a first machine learning model to determine a depth of the water region;
analyzing, by the one or more processors, the one or more images using a second machine learning model to determine a surface normal of the water region;
performing, by the one or more processors, using a third machine learning model, multi-class segmentation of the one or more images; and
performing one or more fusion operations on outputs of at least two of the first machine learning model, the second machine learning model and the third machine learning model to generate a classification of the water region.
27 . The device of claim 26 , wherein the one or more means comprise one or more processors implemented in circuitry.
28 . The device of claim 27 , further comprising a memory to store the one or more images.
29 . The device of claim 26 , further comprising a display configured to display the classification of the water region.
30 . The device of claim 26 , wherein the device comprises one or more of a camera, a computer, a mobile device, a broadcast receiver device, or a set-top box.