STEREOSCOPIC PEDESTRIAN DETECTION SYSTEM WITH TWO-STREAM NEURAL NETWORK WITH DEEP TRAINING AND METHODS OF APPLICATION THEREOF Russian patent published in 2020 - IPC G06K9/00 G06N3/08 G06N3/04 H04N13/239 G06K9/62 G06T7/593 

FIELD: data processing.

SUBSTANCE: invention relates to processing video information. Pedestrian detection system comprises a stereoscopic chamber for capturing certain stereoscopic images of pedestrians passing through a predetermined portion, an ASIC circuit for processing captured stereoscopic images, a stereoscopic pedestrian detection system controller having a processor, a network interface and memory in which computer-executable instructions are stored which cause the processor to: capturing stereoscopic stereoscopic images of pedestrians, rectification of stereoscopic images, calculating disparity maps of rectified stereoscopic images, training a double-stream neural network with deep training, which comprises a neural network for deriving disparities from disparity maps of a plurality of stereoscopic pedestrian images, and a neural network for studying and combining features derived from left rectified images and disparities maps of a plurality of stereoscopic pedestrian images, detecting a plurality of pedestrians passing through a predetermined portion by a trained double-flow neural network with deep training.

EFFECT: enabling creation of a stereoscopic pedestrian detection system.

20 cl, 9 dwg