The project deals with research and development of technologies for intelligent optical tracking system, which will allow the detection, identification and tracking of several objects, located at the same time in the field of view of an optical system.
RCE team is concerned with the problem of tracking generic long distance targets in real time using an active camera. Due to color information loss in very distant objects, we especially focus on tracking issues in monochromatic video sequences using particle filter. The presented solution is capable of processing more than 20 frames per second and it is robust to both partial and full occlusions. This performance has been achieved by utilizing a highly parallel architecture (NVidia graphic card) for the computation. All parts of the mechatronic system can be controlled remotely via ONVIF protocol. The video captured by the high-resolution camera is encoded to H.264 elementary video format and streamed to a connected client via RTP/RTSP protocol.
Video
Hardware
Monochromatic high resolution camera Prosilica GC1290 is mounted on electronically controlled manipulator MST01. Acquired camera images are processed by a high performance personal computer composed of m-ITX motherboard ASUS with four-core Intel processor i5, 4 GB operating memory and GTX 560 Ti NVidia graphics card with 384 CUDA cores.
Software
The control software is a modular multithread application with graphic interface. The software itself is mostly realized in C++. Parts of the software critical for the speed of computation have been hand-optimized for parallel processing on both the processor and the graphic card.
Participants
This device has been realized under the project Research and development of technologies for intelligent optical tracking systems (FR-TI1/195). The main contractor was the company Oprox, a.s., other participants were Faculty of Information Technology, BUT and ZLD, s.r.o.