Seatbelt detection using computer vision
An apparatus includes a capture device and a processor. The capture device may be configured to generate a plurality of video frames corresponding to an interior view of a vehicle. The processor may be configured to perform operations to detect objects in the video frames, detect (a) occupants of the vehicle and (b) seats of the vehicle based on the objects detected in the video frames, determine a status of a seatbelt for each of the occupants and select a reaction based on (a) a state of the seatbelt and (b) characteristics of the occupants. The reaction may be selected to encourage proper usage of the seatbelt based on the characteristics of the occupants. The characteristics may be determined by performing the operations on each of the occupants.
1. An apparatus comprising:
an interface configured to receive pixel data corresponding to an interior view of a vehicle; and
a processor configured to
(i) generate video frames from said pixel data,
(ii) perform computer vision operations to detect objects in said video frames,
(iii) detect (a) occupants of said vehicle, (b) seats of said vehicle, and (c) a status of a seatbelt for each of said occupants, each based on said objects detected in said video frames, and
(iv) select a reaction based on (a) said status of said seatbelt and (b) characteristics of said occupants, wherein (a) said characteristics are determined by performing said computer vision operations on each of said occupants, (b) said status of said seatbelt is determined in response to a relationship between said characteristics of said occupants relative to said seats, (c) said computer vision operations detect said objects by performing feature extraction based on neural network weight values for each of a plurality of visual features that are associated with said objects extracted from said video frames and (d) said neural network weight values are determined in response to an analysis of training data by said processor prior to said feature extraction.
2. The apparatus according to claim 1 , wherein said reaction implements an automatic adjustment of said seatbelt based on said characteristics of said occupants determined using said computer vision operations.
3. The apparatus according to claim 2 , wherein (i) said processor is configured to generate a signal to activate a motor and (ii) said motor is configured to perform said reaction by adjusting a shoulder hinge of said seatbelt.
4. The apparatus according to claim 1 , wherein said computer vision operations are implemented by a convolutional neural network.
5. The apparatus according to claim 4 , wherein said convolutional neural network is trained using fleet learning.
6. The apparatus according to claim 5 , wherein (i) said fleet learning comprises capturing reference images using a capture device in a vehicle production facility, (ii) said reference images comprise an unoccupied interior of a vehicle, (iii) said reference images are used as said training data for said convolutional neural network and (iv) said training data comprises said reference images from many different vehicles.
7. The apparatus according to claim 6 , wherein (i) said training data is uploaded to a central source for training said convolutional neural network and (ii) results of said training are installed on said processor.
8. The apparatus according to claim 1 , wherein said reaction comprises an audible warning.
9. The apparatus according to claim 1 , wherein said characteristics comprise a location of shoulders of said occupants.
10. The apparatus according to claim 1 , wherein said characteristics comprise a size of said occupants.
11. The apparatus according to claim 1 , wherein said status of said seatbelt comprises whether said seatbelt is properly worn by said occupants.
12. The apparatus according to claim 1 , wherein said computer vision operations are configured to detect a sequence of movements by said occupants.
13. The apparatus according to claim 12 , wherein said reaction is not performed when said sequence of movements is determined to be a person fastening said seatbelt.
14. The apparatus according to claim 1 , wherein said processor has a plurality of co-processors.
15. The apparatus according to claim 1 , wherein (i) a capture device configured to generate said pixel data comprises a stereo camera pair and (ii) said computer vision operations comprise performing stereo vision to determine depth information based on said video frames captured by said stereo camera pair.
16. The apparatus according to claim 1 , wherein said computer vision operations are further performed by (i) applying a feature detection window to each of a plurality of layers extracted from said video frames and (ii) a convolution operation using matrix multiplication of said plurality of layers defined by said feature detection window.
17. The apparatus according to claim 16 , wherein said computer vision operations are further performed by sliding said feature detection window along each of said plurality of layers.
18. The apparatus according to claim 1 , wherein (i) said reaction is selected to encourage proper usage of said seatbelt based on said characteristics of said occupants and (ii) said proper usage of said seatbelt comprises determining whether said characteristics of said occupants and said status of said seatbelt comply with regulations.
19. The apparatus according to claim 1 , wherein said relationship between said characteristics of said occupants relative to said seats is determined in response to using said computer vision operations to identify body parts of said occupants and measure a distance between said body parts to determine a size of said occupants relative to said seats.
20. An apparatus comprising:
an interface configured to receive pixel data corresponding to an interior view of a vehicle; and
a processor configured to
(i) generate video frames from said pixel data,
(ii) perform computer vision operations to detect objects in said video frames,
(iii) detect (a) occupants of said vehicle, (b) seats of said vehicle, and (c) a status of a seatbelt for each of said occupants, each based on said objects detected in said video frames, and
(iv) select a reaction based on said status of said seatbelt, wherein (a) said computer vision operations detect said objects by performing feature extraction based on neural network weight values for each of a plurality of visual features that are associated with said objects extracted from said video frames and (b) said neural network weight values are determined in response to an analysis of training data by said processor prior to said feature extraction.