Enhanced face-detection and face-tracking for embedded vision systems

Abstract

Embodiments described herein provide various examples of a face-detection system. In one aspect, a process for performing image detections on grayscale images is disclosed. This process can begin by receiving a training image dataset, wherein the training image dataset includes a first subset of color images. The process then converts each image in the first subset of color images in the training image dataset into a grayscale image to obtain a first subset of converted grayscale images. Next, the process trains an image-detection statistical model using the training image dataset including the first subset of converted grayscale images. The process next receives a set of grayscale input images. The process subsequently performs image detections on the set of grayscale input images using the trained image-detection statistical model. Note that performing image detections on grayscale input images using an image-detection model trained on grayscale training images improves image detection accuracy over using an image-detection model trained on color training images.

What is claimed is: 
     
         1 . A computer-implemented method for performing image detections, the method comprising:
 receiving a training image dataset, wherein the training image dataset includes a first subset of color images;   converting each image in the first subset of color images in the training image dataset into a grayscale image to obtain a first subset of converted grayscale images;   training an image-detection statistical model using the training image dataset including the first subset of converted grayscale images;   receiving a set of grayscale input images; and   performing image detections on the set of grayscale input images using the trained image-detection statistical model.   
     
     
         2 . The computer-implemented method of  claim 1 , wherein the training image dataset is a large-scale public training dataset composed of primarily color images. 
     
     
         3 . The computer-implemented method of  claim 1 , wherein the set of grayscale input images are captured under a monochrome or a grayscale illumination condition. 
     
     
         4 . The computer-implemented method of  claim 3 , wherein the monochrome or grayscale illumination condition includes an LED lighting. 
     
     
         5 . The computer-implemented method of  claim 1 , wherein the set of grayscale input images are captured by a camera configured to capture only grayscale images. 
     
     
         6 . The computer-implemented method of  claim 1 , wherein the training image dataset further includes a second subset of grayscale images, and wherein training the image detection statistical model using the training image dataset includes using both the first subset of converted grayscale images and the second subset of grayscale images. 
     
     
         7 . The computer-implemented method of  claim 1 , wherein the training image dataset is a face-image training dataset, and wherein the image-detection statistical model is a face-detection statistical model. 
     
     
         8 . The computer-implemented method of  claim 7 , wherein the face-detection statistical model includes a convolutional-neural-network (CNN) face-detection module, and wherein the CNN face-detection module further includes a multitask-cascaded-CNN (MTCNN). 
     
     
         9 . The computer-implemented method of  claim 1 , wherein converting the training image dataset into grayscale images reduces data distribution skews between the training image dataset and the set of grayscale input images. 
     
     
         10 . The computer-implemented method of  claim 1 , wherein performing image detections on the set of grayscale input images using the image-detection statistical model trained on the grayscale training images improves image detection accuracy over using an image-detection statistical model trained on color training images. 
     
     
         11 . An apparatus for performing image detections, comprising:
 one or more processors; and   a memory coupled to the one or more processors, wherein the memory stores instructions that, when executed by the one or more processors, cause the apparatus to:
 receive a training image dataset, wherein the training image dataset includes a first subset of color images; 
 convert each image in the first subset of color images in the training image dataset into a grayscale image to obtain a first subset of converted grayscale images; 
 train an image-detection statistical model using the training image dataset including the first subset of converted grayscale images; 
 receive a set of grayscale input images; and 
 perform image detections on the set of grayscale input images using the trained image-detection statistical model. 
   
     
     
         12 . The apparatus of  claim 11 , wherein the training image dataset is a large-scale public training dataset composed of primarily color images. 
     
     
         13 . The apparatus of  claim 11 , wherein the set of grayscale input images are captured under a monochrome or a grayscale illumination condition. 
     
     
         14 . The apparatus of  claim 13 , wherein the monochrome or grayscale illumination condition includes an LED lighting. 
     
     
         15 . The apparatus of  claim 11 , wherein the set of grayscale input images are captured by a camera configured to capture only grayscale images. 
     
     
         16 . The apparatus of  claim 11 , wherein the training image dataset further includes a second subset of grayscale images, and wherein training the image detection statistical model using the training image dataset includes using both the first subset of converted grayscale images and the second subset of grayscale images. 
     
     
         17 . The apparatus of  claim 11 , wherein the training image dataset is a face-image training dataset, and wherein the image-detection statistical model is a face-detection statistical model. 
     
     
         18 . The apparatus of  claim 17 , wherein the face-detection statistical model includes a convolutional-neural-network (CNN) face-detection module, and wherein the CNN face-detection module further includes a multitask-cascaded-CNN (MTCNN). 
     
     
         19 . The embedded system of  claim 14 , wherein performing image detections on the set of grayscale input images using the image-detection statistical model trained on the grayscale training images improves image detection accuracy over using an image-detection statistical model trained on color training images. 
     
     
         20 . A system for performing image detections, comprising:
 a machine learning module comprising a non-transitory computer-readable medium comprising instructions that, when executed by one or more processors, cause the system to to:
 receive a training image dataset, wherein the training image dataset includes a first subset of color images; 
 convert each image in the first subset of color images in the training image dataset into a grayscale image to obtain a first subset of converted grayscale images; 
 train an image-detection statistical model using the training image dataset including the first subset of converted grayscale images; 
 receive a set of grayscale input images; and 
 perform image detections on the set of grayscale input images using the trained image-detection statistical model.