Computer software module arrangement, a circuitry arrangement, an arrangement and a method for improved object detection
An object detection arrangement having a controller configured to: a) receive a plurality of image data streams; b) perform feature extraction on each of the received a plurality of images providing a plurality of feature data streams; and to c) perform a common feature extraction based on the plurality of feature data streams providing as common feature data stream for object detection.
1. An object detection arrangement comprising:
a plurality of image capturing devices; and
a controller configured to:
receive a plurality of image data streams, each of the plurality of image data streams originating from one of the plurality of image capturing devices;
perform local feature extraction on each of the received plurality of images to provide a plurality of feature data streams;
perform a common feature extraction at least by applying a plurality of neural network layers exhibiting temporal dynamic behavior to the plurality of feature data streams, the plurality of neural network layers being arranged such that one of the plurality of neural network layers receives input from another of the plurality of neural network layers not immediately preceding the one of the plurality of neural network layers, the common feature extraction being based on the plurality of feature data streams to provide a common feature data stream for object detection; and
at least one of the plurality of image capturing devices is configured to perform the local feature extraction on a corresponding image data stream.
2. The object detection arrangement according to claim 1 , wherein the controller is further configured to perform object detection based on the common feature data stream.
3. The object detection arrangement according to claim 2 , wherein the object detection is based on a Deep Neural Network model.
4. The object detection arrangement according to claim 1 , wherein at least one of the plurality of neural network layers exhibiting temporal dynamic behavior is a recurrent neural network (RNN).
5. The object detection arrangement according to claim 1 , wherein one image data stream comprises images at least partially overlapping images in another image data stream.
6. The object detection arrangement according to claim 1 , wherein the controller is further configured to combine the plurality of feature data streams into a combined feature data stream.
7. The object detection arrangement according to claim 1 , wherein the object detection arrangement is a one of smartphone and a tablet computer.
8. The object detection arrangement according to claim 1 , wherein the object detection arrangement is an optical see-through device.
9. The object detection arrangement according to claim 1 , wherein the object detection arrangement is arranged to be used in image retrieval, industrial use, robotic vision and/or video surveillance.
10. The object detection arrangement according to claim 2 , wherein at least one of the plurality of neural network layers exhibiting temporal dynamic behavior is a recurrent neural network (RNN).
11. The object detection arrangement according to claim 2 , wherein one image data stream comprises images at least partially overlapping images in another image data stream.
12. The object detection arrangement according to claim 2 , further comprising a plurality of image capturing devices, wherein each image data stream originates from one of the image capturing devices.
13. A method for object detection in an object detection arrangement, the method comprising:
receiving a plurality of image data streams, each of the plurality of image data streams originating from one of a plurality of image capturing devices;
performing local feature extraction on each of the received plurality of images to provide a plurality of feature data streams;
performing a common feature extraction at least by applying a plurality of neural network layers exhibiting temporal dynamic behavior to the plurality of feature data streams, the plurality of neural network layers being arranged such that one of the plurality of neural network layers receives input from another of the plurality of neural network layers not immediately preceding the one of the plurality of neural network layers, the common feature extraction being based on the plurality of feature data streams providing to provide a common feature data stream for object detection; and
configuring at least one of the plurality of image capturing devices to perform the local feature extraction on a corresponding image data stream.
14. A non-transitory computer-readable storage medium storing computer instructions that when loaded into and executed by a controller of an object detection arrangement causes the object detection arrangement to perform a method, the method comprising:
receiving a plurality of image data streams, each of the plurality of image data streams originating from one of a plurality of image capturing devices;
performing local feature extraction on each of the received plurality of images to provide a plurality of feature data streams;
performing a common feature extraction at least by applying a plurality of neural network layers exhibiting temporal dynamic behavior to the plurality of feature data streams, the plurality of neural network layers being arranged such that one of the plurality of neural network layers receives input from another of the plurality of neural network layers not immediately preceding the one of the plurality of neural network layers, the common feature extraction being based on the plurality of feature data streams to provide a common feature data stream for object detection; and
configuring at least one of the plurality of image capturing devices to perform the local feature extraction on a corresponding image data stream.