Large animal detection and intervention in a vehicle
A far-infrared camera mounted in a vehicle generates an image frame. When an image of a large animal is identified in the image frame, a pixel intensity of the large animal image is determined. An estimated distance to the large animal from the far-infrared camera based on the pixel intensity is determined. When the animal is classified as a tracked animal, and future trajectories of the tracked animal and the vehicle intersect, a component in the vehicle is actuated.
1 . A computing device, comprising a processor and a memory, wherein the memory stores instructions executable by the processor such that the computing device is programmed to:
upon identifying an image of a large animal in an image frame obtained with a far-infrared camera mounted to a vehicle, wherein the identifying includes identifying a type of the large animal selected from a group including one or more of deer, bear, elk, horse, or cow:
determine a pixel intensity of the large animal image;
determine an estimated distance to the large animal from the far-infrared camera based on the pixel intensity;
classify the large animal as one of a tracked animal and a non-tracked animal based on the large animal image, including a pose based on the type of the large animal and whether the pose indicates that the large animal is likely to occupy a road; and
upon classifying the large animal as the tracked animal and upon determining that a future trajectory of the tracked animal and a future trajectory of the vehicle intersect, actuate a component in the vehicle.
2 . The computing device of claim 1 , wherein the computing device is further programmed to:
generate a sequence of image frames; and
predict the future trajectory of the tracked animal based on a past trajectory of the tracked animal, a scene context, and a pixel motion.
3 . The computing device of claim 1 , wherein the computing device is further programmed to generate a bounding box that bounds the large animal image.
4 . The computing device of claim 3 , wherein the computing device is further programmed to, prior to classifying the large animal, determine the estimated distance based on both the pixel intensity and on a size of the bounding box.
5 . The computing device of claim 3 , wherein the computing device is further programmed to:
generate a sequence of image frames; and
determine a past trajectory of the tracked animal based on a movement of the bounding box of the large animal image in the sequence of the image frames.
6 . The computing device of claim 1 , wherein the computing device is further programmed to:
generate a sequence of image frames; and
determine the future trajectory of the tracked animal based on a motion of pixels in the sequence of the image frames.
7 . The computing device of claim 1 , wherein the computing device is further programmed to:
generate a sequence of image frames; and
predict the future trajectory of the tracked animal by modeling the future trajectory with a convolutional long short term memory attention-based multi-stream encoder-decoder model that includes an attention module that assigns weights to vectors derived from a past trajectory of the tracked animal, a scene context, and a pixel motion.
8 . The computing device of claim 1 , wherein the component actuated comprises a heads-up display that displays a visual notification of the tracked animal to a vehicle operator.
9 . The computing device of claim 1 , wherein the computing device is further programmed to, prior to classifying the large animal, identify key points of the large animal image and their respective locations in the image frames to determine a pose of the large animal and include the pose as a factor in classifying the large animal.
10 . A method of predicting a future trajectory of a large animal and actuating a component, the method comprising the steps of:
upon identifying an image of a large animal in an image frame obtained with a far-infrared camera mounted to a vehicle, wherein the identifying includes identifying a type of the large animal selected from a group including one or more of deer, bear, elk, horse, or cow:
determining a pixel intensity of the large animal image;
determining an estimated distance to the large animal from the far-infrared camera based on the pixel intensity;
classifying the large animal is classified as one of a tracked animal and a non-tracked animal based on the large animal image, including a pose based on the type of the large animal and whether the pose indicates that the large animal is likely to occupy a road; and
upon classifying the large animal as a tracked animal and upon determining that the future trajectory of the tracked animal and the future trajectory of the vehicle intersect, actuating a component in the vehicle.
11 . The method of claim 10 , further comprising the steps of:
generating a sequence of image frames; and
predicting the future trajectory of the tracked animal based on a past trajectory of the tracked animal, a scene context, and a pixel motion.
12 . The method of claim 10 , further comprising the step of generating a bounding box that bounds the large animal image.
13 . The method of claim 12 , further comprising the step of, prior to classifying the large animal, determining the estimated distance based on both the pixel intensity and on a size of the bounding box.
14 . The method of claim 12 , further comprising the steps of:
generating a sequence of image frames; and
determining a past trajectory of the tracked animal based on a movement of the bounding box of the large animal image in the sequence of the image frames.
15 . The method of claim 10 , further comprising the steps of:
generating a sequence of image frames; and
determining the future trajectory of the tracked animal based on a motion of pixels in the sequence of the image frames.
16 . The method of claim 10 , further comprising the steps of:
generating a sequence of image frames; and
predicting the future trajectory of the tracked animal by modeling the future trajectory with a convolutional long short term memory attention-based multi-stream encoder-decoder model that includes an attention module and assigning weights to vectors derived from a past trajectory of the tracked animal, a scene context, and a pixel motion.
17 . The method of claim 10 , wherein the component being actuated comprises a heads-up display that displays a visual notification of the tracked animal to a vehicle operator.
18 . The method of claim 10 , further comprising the step of, prior to classifying the large animal, identifying key points of the large animal image and their respective locations in the image frames to determine a pose of the large animal and including the pose as a factor in classifying the large animal.