Retinal encoder for machine vision
A method is disclosed including: receiving raw image data corresponding to a series of raw images; processing the raw image data with an encoder to generate encoded data, where the encoder is characterized by an input/output transformation that substantially mimics the input/output transformation of one or more retinal cells of a vertebrate retina; and applying a first machine vision algorithm to data generated based at least in part on the encoded data.
1. A method including:
applying, by an encoding module, a spatiotemporal transformation to image data to generate retinal output cell response values;
generating, by the encoding module, encoded data based on the retinal output cell response values;
applying, by a machine vision module, a machine vision algorithm to the encoded data;
monitoring, by a controller, performance of the machine vision algorithm, wherein monitoring the performance of the machine vision algorithm comprises:
calculating an error rate of the machine vision algorithm; and
comparing the error rate to a threshold level; and
adjusting, by the controller, the machine vision algorithm based on the monitored performance.
2. The method of claim 1 , further comprising generating retinal images from the retinal output cell response values, and wherein applying the machine vision algorithm to the encoded data comprises applying the machine vision algorithm to the retinal images.
3. The method of claim 2 , further comprising determining pixel values in the retinal images based on the encoded data, where determining pixel values in the retinal images includes determining a pixel intensity or color indicative of a retinal cell response, and wherein the data indicative of a retinal cell response is indicative of at least one of a retinal cell firing rate, a retinal cell output pulse train, and a generator potential.
4. The method of claim 1 , wherein the machine vision algorithm includes at least one selected from the list consisting of: an object recognition algorithm, an image classification algorithm, a facial recognition algorithm, an optical character recognition algorithm, a content-based image retrieval algorithm, a pose estimation algorithm, a motion analysis algorithm, an egomotion determination algorithm, a movement tracking algorithm, an optical flow determination algorithm, a scene reconstruction algorithm, a 3D volume recognition algorithm, and a navigation algorithm.
5. The method of claim 1 , wherein adjusting the machine vision algorithm is responsive to a determination that the error rate exceeds the threshold level.
6. The method of claim 5 , wherein adjusting the machine vision algorithm comprises iteratively adjusting one or more parameters until the error rate of the machine vision algorithm satisfies the threshold level.
7. The method of claim 1 , wherein adjusting the machine vision algorithm comprises modifying a parameter of the machine vision algorithm.
8. The method of claim 1 , where applying the machine vision algorithm comprises applying a navigation algorithm, where applying the navigation algorithm includes:
processing the encoded data to determine motion information indicative of motion at a plurality of image locations in the encoded data;
classifying spatial regions in the encoded data based on the motion information; and
generating a navigation decision based on the classification of the spatial regions.
9. The method of claim 8 , wherein the motion information is indicative of an optical flow in the encoded data, the method further including:
using a convolutional neural network to classify the spatial regions; and
controlling the motion of a robotic apparatus based on results from navigation algorithm.
10. A method including:
applying, by an encoding module, a spatiotemporal transformation to image data to generate retinal output cell response values;
generating, by the encoding module, encoded data based on the retinal output cell response values;
applying, by a machine vision module, a machine vision algorithm to the encoded data;
monitoring, by a controller, performance of the machine vision algorithm, wherein monitoring the performance of the machine vision algorithm comprises:
calculating an error rate of the machine vision algorithm; and
comparing the error rate to a threshold level; and
adjusting, by the controller, the machine vision algorithm based on the monitored performance, wherein adjusting the machine vision algorithm comprises modifying a parameter of the machine vision algorithm, and wherein the machine vision algorithm comprises an artificial neural network, and wherein adjusting the machine vision algorithm comprises changing a plurality of connections in an artificial neural network.
11. A method including:
applying, by an encoding module, a spatiotemporal transformation to image data to generate retinal output cell response values;
generating, by the encoding module, encoded data based on the retinal output cell response values;
applying, by a machine vision module, a machine vision algorithm to the encoded data:
monitoring, by a controller, performance of the machine vision algorithm, wherein monitoring the performance of the machine vision algorithm comprises:
calculating an error rate of the machine vision algorithm; and
comparing the error rate to a threshold level; and
adjusting, by the controller, the machine vision algorithm based on the monitored performance, wherein adjusting the machine vision algorithm comprises iteratively adjusting one or more parameters of the machine vision algorithm until an incremental increase in performance per iteration falls below a threshold level.
12. An apparatus including:
a memory storage device configured to store image data corresponding to a series of images; and
a processor operably coupled with the memory and programmed to:
receive the image data corresponding to the series of images;
generate encoded data from the image data, wherein, to generate the encoded data, the processor is configured to:
apply a spatiotemporal transformation to image data to generate retinal output cell response values;
generate encoded data based on the retinal output cell response values;
apply a machine vision algorithm to the encoded data;
monitor performance of the machine vision algorithm, wherein, to monitor the performance of the machine vision algorithm, the processor is further configured to:
calculate an error rate of the machine vision algorithm; and
compare the error rate to a threshold level; and
adjust the machine vision algorithm based on the monitored performance.
13. The apparatus of claim 12 , wherein the processor is further configured to generate retinal images from the retinal output cell response values, and wherein applying the machine vision algorithm to the encoded data comprises applying the machine vision algorithm to the retinal images.
14. The apparatus of claim 12 , wherein, to adjust the machine vision algorithm, the processor is further configured to iteratively adjust one or more parameters of the machine vision algorithm until an error rate of the machine vision algorithm satisfies the threshold level.
15. The apparatus of claim 12 , wherein, to adjust the machine vision algorithm, the processor is further configured to iteratively adjust one or more parameters of the machine vision algorithm until an incremental increase in performance of the machine vision algorithm per iteration falls below a threshold level.
16. A non-transitory computer-readable medium having computer-executable instructions for implementing operations comprising
applying a spatiotemporal transformation to image data to generate retinal output cell response values;
generating encoded data based on the retinal output cell response values;
applying a machine vision algorithm to the encoded data;
monitoring performance of the machine vision algorithm; and
adjusting the machine vision algorithm based on the monitored performance, wherein adjusting the machine vision algorithm comprises iteratively adjusting one or more parameters of the machine vision algorithm until an incremental increase in performance per iteration of the machine vision algorithm falls below a threshold level.
17. The non-transitory computer-readable medium of claim 16 , wherein monitoring the performance of the machine vision algorithm comprises:
calculating an error rate of the machine vision algorithm; and
comparing the error rate to a threshold level; and
wherein adjusting the machine vision algorithm is responsive to a determination that the error rate exceeds the threshold level.