Dynamic Camera Reconfiguration with Reinforcement Learning and Stochastic Methods for Crowd Surveillance

Abstract

Crowd surveillance plays a key role to ensure safety and security in public areas. Surveillance systems traditionally rely on fixed camera networks, which suffer from limitations, as coverage of the monitored area, video resolution and analytic performance. On the other hand, a smart camera network provides the ability to reconfigure the sensing infrastructure by incorporating active devices such as pan-tilt-zoom (PTZ) cameras and UAV-based cameras, thus enabling the network to adapt over time to changes in the scene. We propose a new decentralised approach for network reconfiguration, where each camera dynamically adapts its parameters and position to optimise scene coverage. Two policies for decentralised camera reconfiguration are presented: a greedy approach and a reinforcement learning approach. In both cases, cameras are able to locally control the state of their neighbourhood and dynamically adjust their position and PTZ parameters. When crowds are present, the network balances between global coverage of the entire scene and high resolution for the crowded areas. We evaluate our approach in a simulated environment monitored with fixed, PTZ and UAV-based cameras.

Eagle: End-to-end Deep Reinforcement Learning based Autonomous Control of PTZ Cameras

“…Object_detection+reinforcement learning [7,13]: Steps 1 2 5 9 . These methods combine the tracking and control stages with a deep-RL policy.…”

“…These methods combine the tracking and control stages with a deep-RL policy. Bisagno et al [7] and Kim et al [13] show the control of PTZ camera using deep-RL where the inputs to the neural networks are the information about the object of interest (e.g., bounding-boxes, location of pedestrians). An actual deployment may need to use external object detectors to measure these inputs, where the performance suffers from object detector errors.…”

“…Autonomous control of pantilt-zoom (PTZ) cameras provides superior monitoring with active vision systems by tracking objects of interest in real-time [9,17]. Active vision systems are increasingly deployed in resource-constrained environments such as remote surveillance [7,11,17] and mobile robotics [9,25].…”

Visual Sensor Networks for Indoor Real-Time Surveillance and Tracking of Multiple Targets

“…For instance, a Q-learning-based solution was presented in [ 17 ] for smoothly controlling a PTZ camera. A soft actor–critic RL system was, instead, designed in [ 18 ] to perform the decentralized reconfiguration of a camera network involving PTZ sensors and devices mounted on flying platforms. The main issue related to the RL solutions consists of their intrinsic black box nature: in some contexts, it is not possible or safe to rely upon such controllers, whose performance may not be completely predictable.…”

Flying Chameleons: A New Concept for Minimum-Deployment, Multiple-Target Tracking Drones

“…In this case, apart from compensating for the tilts and/or turns undergone by the vehicle, it is desired to simultaneously keep track of moving targets [ 2 , 3 , 4 , 5 , 6 ]. The introduction of zoom camera lenses brought superior mission capabilities [ 7 , 8 ]. Having high magnification factors, however, makes fine image stabilization a serious challenge, usually requiring the conjunction of high-precision gimbal control and optical or electronic image stabilization strategies [ 9 , 10 ].…”