System and method for occupancy monitoring

Abstract

There is provided systems and methods for automated water operations for aquatic facilities using at least one image captured of the aquatic facilities. A method includes: receiving an input signal including a detected number of occupants in the water at the aquatic facilities, the number of occupants determined using a trained detection machine learning model, the detection machine learning model receiving the captured image with an associated feature map as input, and outputting a detection of each occupant in the water, the water level machine learning model trained using training images each including a respective label for each occupant in the training image; determining a volume of water to add by multiplying the number of occupants by a predetermined volume of freshwater to add per occupant; and directing one or more water flow regulators to permit inflow of water approximately equivalent to the volume of water to add.

1 . A computer-implemented method for occupancy monitoring of a facility using at least one image captured of the facility, the method comprising:
 receiving the at least one captured image;   receiving an input signal comprising a detected number of occupants in the facility, the number of occupants captured in the at least one captured image determined using a trained detection machine learning model, the detection machine learning model taking as input the at least one captured image with an associated feature map, and outputting a detection of each occupant in the facility, the detection machine learning model trained using training images each comprising a respective label for each occupant in the training image; and   outputting the detected number of occupants.   
     
     
         2 . The method of  claim 1 , further comprising performing semi-supervised background subtraction to remove areas not capturing areas of occupancy from the captured image that is inputted to the trained detection machine learning model. 
     
     
         3 . The method of  claim 2 , wherein the background subtraction comprises separating occupants as foreground elements from the background by generating a foreground mask. 
     
     
         4 . The method of  claim 3 , wherein the foreground elements are determined by detecting dynamically moving objects. 
     
     
         5 . The method of  claim 4 , wherein receiving the at least one captured image comprises receiving multiple successive captured images, and wherein detecting dynamically moving objects comprises determining a running average as a function over the successive captured images. 
     
     
         6 . The method of  claim 5 , wherein the running average is determined using:
   FG( x,y )=CF( x,y )−BG( x,y )
   
       wherein FG are coordinates of foreground elements, CF are coordinates in the current frame, and BG are coordinates in a background model. 
     
     
         7 . The method of  claim 1 , wherein the detection machine learning model comprises a region proposal network. 
     
     
         8 . The method of  claim 7 , wherein the region proposal network comprises a ResNet-50 architecture to extract features of occupants and a fully connected network to localize and classify the occupants using the features. 
     
     
         9 . A system for occupancy monitoring of a facility using at least one image captured of the facility, the system comprising one or more processors and a data storage, the one or more processors configured to execute:
 an input module to receive the at least one captured image from the one or more cameras;   an occupant detection module to:
 receive an input signal comprising a detected number of occupants in the facility, the number of occupants captured in the at least one captured image determined using a trained detection machine learning model, the detection machine learning model taking as input the at least one captured image with an associated feature map, and outputting a detection of each occupant in the facility, the detection machine learning model trained using training images each comprising a respective label for each occupant in the training image; and 
   an output module to output the detected number of occupants.   
     
     
         10 . The system of  claim 9 , wherein the occupant detection module performs semi-supervised background subtraction to remove areas not capturing areas of occupancy from the captured image that is inputted to the trained detection machine learning model. 
     
     
         11 . The system of  claim 10 , wherein the background subtraction comprises separating occupants as foreground elements from the background by generating a foreground mask. 
     
     
         12 . The system of  claim 11 , wherein the foreground elements are determined by detecting dynamically moving objects. 
     
     
         13 . The system of  claim 12 , wherein receiving the at least one captured image comprises receiving multiple successive captured images, and wherein detecting dynamically moving objects comprises determining a running average as a function over the successive captured images. 
     
     
         14 . The system of  claim 13 , wherein the running average is determined using:
   FG( x,y )=CF( x,y )−BG( x,y )
   
       wherein FG are coordinates of foreground elements, CF are coordinates in the current frame, and BG are coordinates in a background model. 
     
     
         15 . The system of  claim 9 , wherein the detection machine learning model comprises a region proposal network. 
     
     
         16 . The system of  claim 15 , wherein the region proposal network comprises a ResNet-50 architecture to extract features of occupants and a fully connected network to localize and classify the occupants using the features.