Pruning filters for efficient convolutional neural networks for image recognition in vehicles
Systems and methods for surveillance are described, including an image capture device configured to mounted to an autonomous vehicle, the image capture device including an image sensor. A storage device is included in communication with the processing system, the storage device including a pruned convolutional neural network (CNN) being trained to recognize obstacles in a road according to images captured by the image sensor by training a CNN with a dataset and removing filters from layers of the CNN that are below a significance threshold for image recognition to produce the pruned CNN. A processing device is configured to recognize the obstacles by analyzing the images captured by the image sensor with the pruned CNN and to predict movement of the obstacles such that the autonomous vehicle automatically and proactively avoids the obstacle according to the recognized obstacle and predicted movement.
1. A sensing system comprising:
an image capture device configured to mounted to an autonomous vehicle, the image capture device including an image sensor;
a storage device in communication with the processing system, the storage device including a pruned convolutional neural network (CNN) being trained to recognize obstacles in a road according to images captured by the image sensor by training a CNN with a dataset, identifying filters from layers of the CNN that have kernel weight sums that are below a significance threshold for image recognition, and removing the identified filters to produce the pruned CNN, wherein the significance threshold is a number of smallest filters of a convolutional layer of the CNN according to corresponding absolute kernel weight sums; and
a processing device configured to recognize the obstacles by analyzing the images captured by the image sensor with the pruned CNN and to predict movement of the obstacles such that the autonomous vehicle automatically and proactively avoids the obstacle according to the recognized obstacle and predicted movement.
2. The system of claim 1 , wherein the image capture device includes a digital camera.
3. The system of claim 1 , further including a system-on-chip including the image sensor, the storage device and the processing device.
4. The system of claim 1 , wherein the storage device and the processing device are embedded in the image capture device.
5. The system of claim 1 , wherein the processing device further includes a long short-term memory network to analyze the recognized obstacles and generate the predicted movement.
6. The system of claim 1 , wherein the processing device is configured to prune filters from the pruned CNN according to the recognized obstacles.
7. The system of claim 1 , wherein the image sensor includes light detection and ranging (LIDAR) sensor.
8. The system of claim 1 , further including a battery to power the sensing system.
9. A sensing system comprising:
an image capture device including an image sensor and configured to be mounted to a vehicle;
a storage device in communication with the processing system, the storage device including a pruned convolutional neural network (CNN) being trained to recognize obstacles in a road according to images captured by the image sensor by training a CNN with a dataset, identifying filters from layers of the CNN that have kernel weight sums that are below a significance threshold for image recognition, and removing the identified filters to produce the pruned CNN, wherein the significance threshold is a number of smallest filters of a convolutional layer of the CNN according to corresponding absolute kernel weight sums;
a processing device configured to recognize the obstacles by analyzing the images captured by the image sensor with the pruned CNN and to predict movement of the obstacles; and
a notification device configured to notify an operator of the predicted movement.
10. The system of claim 9 , wherein the image capture device includes a digital camera.
11. The system of claim 9 , further including a system-on-chip including the image sensor, the storage device and the processing device.
12. The system of claim 9 , wherein the storage device and the processing device are embedded in the image capture device.
13. The system of claim 9 , wherein the image capture device is removably mounted to the vehicle.
14. The system of claim 9 , wherein the processing device is configured to prune filters from the pruned CNN according to the recognized obstacles.
15. The system of claim 9 , wherein the image sensor includes light detection and ranging (LIDAR) sensor.
16. The system of claim 9 , further including a battery to power the sensing system.