FPGA-BASED SYSTEM AND METHOD FOR DYNAMICALLY SWAPPING CNNS TO IMPROVE DETECTION AND CLASSIFICATION
Systems and methods for image recognition using FPGA computer vision devices with a field-programming gate array (FPGA) and multiple convolutional neural networks (CNNs). A first CNN is used for initial object class detection. A second CNN is swapped for the first CNN for more detailed object classification and identification.
1 . A device for capturing and classifying a digital image of an object, the device comprising:
a flash memory;
an image sensor configured for capturing a plurality of image frames of the object;
an image processing module configured to receive the plurality of image frames from the image sensor and to filter, enhance, or scale the plurality image frames;
a first convolutional neural network (CNN) configured for processing the image frames and identifying a coarse object class;
a second CNN configured for detection of a particular subclass, wherein the first and second CNNs are stored in the flash memory;
an FPGA control block, operably connected to the flash memory, the image processing module, and is responsible for executing the first CNN, and the second CNN, the FPGA control block configured to:
direct the first CNN to determine whether the object is within the coarse object class with a predetermined degree of certainty, and
direct the second CNN to process the object and determine whether the object is within a subclass of the coarse object class when the object is determined to be in the coarse object class by the first CNN with the predetermined degree of certainty; and
a transmitter, configured to send a message to a receiver, the message comprising results of the object class determination of the first CNN or the object subclass definition of the second CNN.
2 . The device of claim 1 , wherein the predetermined degree of certainty corresponds to a successful determination that the object is within the coarse object class.
3 . The device of claim 1 , wherein the predetermined degree of certainty corresponds to an uncertain determination that the object is within the coarse object class.
4 . The device of claim 1 , wherein the FPGA control block comprises a register for storing rules for determining the predetermined threshold for directing the second CNN to process image frames of the object.
5 . The device of claim 1 , wherein the FPGA control block comprises a register for storing rules identifying the coarse object classes whose objects the second CNN is permitted to process.
6 . The device of claim 2 , wherein the FPGA control block is further configured to direct the second CNN to process the object when the second CNN has a dataset corresponding to a subclass of the coarse object class.
7 . The device of claim 3 , wherein the FPGA control block is further configured to direct the second CNN to process the object despite a reduction in processing speed of the image processing module.
8 . A method of capturing and classifying a digital image of an object, the method comprising:
capturing a plurality of image frames of the object using an image sensor:
filtering, enhancing, or scaling the plurality of image frames using an image processing module;
processing the image frames using a first convolutional neural network (CNN) under the control of a Field-Programmable gate array (FPGA) control block to identify a coarse object class, determine a probability of recognition of the object, pose estimation, and determine coordinates of the object;
determining whether the object is within the coarse object class with a predetermined degree of certainty;
when the object is determined to be in the coarse object class with the predetermined degree of certainty, processing the object using a second CNN under the control of the FPGA control block to determine whether the object is within a subclass of the coarse object class, determine a probability of recognition of the object, pose estimation, and determine coordinates of the object; and
sending a message to a receiver, the message comprising at least one of the coarse object class, the subclass of the coarse object class determined by CNNs, the probability of recognition of the object, the pose estimation, and the coordinates of the object within the image assumed by the CNNs.
9 . The method of claim 8 , wherein the predetermined degree of certainty corresponds to a successful determination that the object is within the coarse object class.
10 . The method of claim 8 , wherein the predetermined degree of certainty corresponds to an uncertain determination that the object is within the coarse object class.
11 . The method of claim 8 , wherein the FPGA control block comprises a register for storing rules identifying the coarse object classes whose objects the second CNN is permitted to process.
12 . The method of claim 8 , wherein the FPGA control block is further configured to direct the second CNN to process the object when the second CNN has a dataset corresponding to a subset of the coarse object class.
13 . The method of claim 8 , wherein the FPGA control block is further configured to direct the second CNN to process the object despite a reduction in processing speed of the image processing module.
14 . A method of capturing and classifying a digital image of an object using image frames captured by an image sensor installed at a first location, the method comprising:
capturing a plurality of image frames of the object at the first location using the image sensor;
filtering, enhancing, or scaling the plurality of image frames using an image processing module at the first location;
processing the image frames using a first convolutional neural network (CNN) to identify a coarse object class at the first location;
determining whether the object is within the coarse object class with a predetermined degree of certainty at the first location;
when the object is determined to be in the coarse object class with the predetermined degree of certainty, processing the object using a second CNN to determine whether the object is within a subclass of the coarse object class, determine a probability of recognition of the object, determine pose estimation, and determine coordinates of the object; and
sending a message with one or more of the determinations made by the CNNs to a receiver at a second location, remote from the first location.
15 . The method of claim 14 , wherein the message comprises at least one of the subclass of the coarse object class determined by the second CNN, the probability of recognition, the pose estimation and the coordinates of the object within the image determined by the second CNN.
16 . The method of claim 14 , wherein the predetermined degree of certainty corresponds to a successful determination that the object is within the coarse object class.
17 . The method of claim 16 , wherein the uncertain determination that the object is within the coarse object class results from environmental conditions at the first location.
18 . The method of claim 14 , wherein the FPGA control block is further configured to direct the second CNN to process the object when the second CNN has a dataset corresponding to a subset of the coarse object class.
19 . The method of claim 14 , wherein the FPGA control block is further configured to direct the second CNN to process the object despite a reduction in processing speed of the image processing module.
20 . The method of claim 14 , wherein the predetermined degree of certainty is adjusted dynamically based on available system resources.