US 11,055,872 B1
Real-time object recognition using cascaded features, deep learning and multi-target tracking
Yang Chen, Westlake Village, CA (US); Deepak Khosla, Camarillo, CA (US); and Ryan M. Uhlenbrock, Calabasas, CA (US)
Assigned to HRL Laboratories, LLC, Malibu, CA (US)
Filed by HRL Laboratories, LLC, Malibu, CA (US)
Filed on Jan. 30, 2018, as Appl. No. 15/883,822.
Claims priority of provisional application 62/479,204, filed on Mar. 30, 2017.
Int. Cl. G06T 7/73 (2017.01); G06T 7/246 (2017.01); G06N 3/02 (2006.01)
CPC G06T 7/73 (2017.01) [G06N 3/02 (2013.01); G06T 7/246 (2017.01); G06T 2207/20081 (2013.01)] 20 Claims
OG exemplary drawing
1. An object recognition system, the system comprising:
one or more processors and a non-transitory computer-readable medium having executable instructions encoded thereon such that when executed, the one or more processors perform operations of:
utilizing a three-stage cascaded classifier for classifying an object, wherein utilizing the three-stage cascaded classifier comprises:
training an integral channel feature (ICF) detector offline using decision trees and a boosted learning approach;
applying the trained ICF detector to an input image of a scene in a first stage of the cascaded classifier to learn a plurality of candidate target regions, wherein each candidate target region represents a candidate object,
wherein the ICF detectors use at least an oriented edges channel and a magnitude of gradients channel, wherein each channel is a sum of pixel values over a rectangular region within the candidate target region;
in a second stage of the cascaded classifier, applying a trained convolutional neural network (CNN) classifier to the input image of the scene and the candidate target regions, resulting in a plurality of initially classified objects; and
using a multi-target Kalman filter tracker in a third stage of the cascaded classifier, tracking the initially classified objects, resulting in a classification output; and
controlling a device based on the classification output.