Nec Corporation

Sibendu Paul

Srimat T Chakradhar

Giuseppe Coviello

Kunal Rao

Murugan Sankaradas

Oliver Po

Utsav Drolia

Systems and methods are provided for dynamically tuning camera parameters in a video analytics system to optimize analytics accuracy. A camera captures a current scene, and optimal camera parameter settings are learned and identified for the current scene using a Reinforcement Learning (RL) engine. The learning includes defining a state within the RL engine as a tuple of two vectors: a first representing current camera parameter values and a second representing measured values of frames of the current scene. Quality of frames is estimated using a quality estimator, and camera parameters are adjusted based on the quality estimator and the RL engine for optimization. Effectiveness of tuning is determined using perceptual Image Quality Assessment (IQA) to quantify a quality measure. Camera parameters are adaptively tuned in real-time based on learned optimal camera parameter settings, state, quality measure, and set of actions, to optimize the analytics accuracy for video analytics tasks.

Adaptive perceptual quality based camera tuning using reinforcement learning

A method for automatically adjusting camera parameters to improve video analytics accuracy during continuously changing environmental conditions is presented. The method includes capturing a video stream from a plurality of cameras, performing video analytics tasks on the video stream, the video analytics tasks defined as analytics units (AUs), applying image processing to the video stream to obtain processed frames, filtering the processed frames through a filter to discard low-quality frames and dynamically fine-tuning parameters of the plurality of cameras. The fine-tuning includes passing the filtered frames to an AU-specific proxy quality evaluator, employing State-Action-Reward-State-Action (SARSA) reinforcement learning (RL) computations to automatically fine-tune the parameters of the plurality of cameras, and based on the reinforcement computations, applying a new policy for an agent to take actions and learn to maximize a reward.

Reinforcement-learning based system for camera parameter tuning to improve analytics

Methods and systems for image filtering include detecting a distortion level of input images, using a distortion detection model that is trained using confidence values generated by a pre-trained image classifier with a set of distorted training images. An analysis is performed on input images having a detected distortion level that is lower than a threshold, with input images having an above-threshold detected distortion level being filtered out.

Growing community of inventors

West Lafayette, IN, United States of America

Sibendu Paul