Real-time detection of moving objects in a dynamic scene from moving robotic vehicles

Abstract

Dynamic scene perception is currently limited to detection ofmoving objectsfrom a static plaform or scenes with flat backgrounds. We discuss novel real-time methods to segment moving objects in the motionfieldformed by a moving camerahobotic plaform with mostly translational motion. Our solution does not erplicitly require any egomotion knowledge, thereby making the solution opplicoble to mobile outdoor robot problems where no IMU information is mailable. We address two problems in dynomic scene perception on the move, first using only 2D monoculor grayscale images, and second where 3D range information from stereo is olso available. Our solution involves real-time optical flow computations, followed by optical jlow field preprocessing to highlight moving object boundaries. In the case where range data from stereo is computed, a 3 0 opticalflow field is estimated by combining range information with 2 0 optical flow estimates, followed by a similar 3 0 flow field preprocessing step. A segmentation of the jlow field using fastfloodfilling then identifies eveiy moving object in the scene with a unique lobel. This novel algorithm is apected to be the critical first step in robust recognition of moving vehicles ondpeople from mobile outdoor robots, and therefore offers a robust solution to the problem of dynamic scene perception in the presence of certain kinds of robot motion. It is envisioned that our algorithm will benefit robot scene perception in urban environments for scient&, commercial and defense applications. Results of our real-time algorithm (see video) on a mobile robot in a scene with a single moving vehicle ore presented.

Fast Object Hypotheses Generation Using 3D Position and 3D Motion

“…The same result was obtained by [12], where dynamic objects are also assumed to move mainly parallel to the image plane and 2D projection of the motion component W in Eqs. (4-5) was imposed being approximately constant, i.e.…”

“…(4-5) was imposed being approximately constant, i.e. xW = yW = k. Talukder et al have also implemented a test based on the constraints (8)(9) to detect independently moving objects in the scene [12].…”

Real-time quasi dense two-frames depth map for Autonomous Guided Vehicles

“…Extracting the structural information from the resulting tensors is done by decomposing them according to (5). The saliency measures λ 1 − λ 2 and λ 2 provide the preferred orientation of each tensor and so, the geometrical element it belongs to.…”

“…On the other hand, video sensors appear well adapted for automotive applications, due to the rich 2-D information contained in a single image and the 3-D information inferred by two. To date, research has mostly focused on the perception from fix and mobile cameras in static or constrained dynamic scenes [5], [7]. Recovering the 3-D from a monocular sensor is considered as an highly computational task.…”

Sensorized robotic sphere for large exterior critical infrastructures supervision

“…From the late 1980s, when the IRIS (industrial remote inspection system) robot performed inspection tasks in nuclear plants, 2,3 to current commercial robotic surveillance solutions (e.g., the patrolling of South Korean's Pohang prison by Roboguard 4 ), security robots have been successfully deployed. In this context, in the most advanced sensing systems, not only ground robots [4][5][6] but unmanned aerial vehicles (UAVs), have been incorporated as part of security and surveillance multirobot systems (MRSs). 7,8,9 Examples of these platforms are the AirRobot's AR100-B (http://www.airrobot.com/index.php/products-28.html) and Astec's Pelican (http://www.asctec.de/asctec-pelican-3/) UAVs.…”