| Object tracking based on computer vision is one of the hotspot research issues in recent years. In a number of fields, such as smart human-computer interaction, smart surveillance, military purposes, video compression, medical diagnosis, intelligent transportation, smart wear, virtual reality and 3D reality, object tracking has very important usage. Improving the precision of object tracking and real-time performance has always been a pursuit.In this paper, the Camshift tracking algorithm and the tracking method based on SURF features have been researched and improved, also validated by designed experiments. In order to solve the wrong tracking which comes up when the color of target and background are similar, the paper proposes a method to detect the wrong tracking by calculating the Bhattacharyya Distance of the color histogram probability. With regard to SURF features, it is easy to get wrong matches between the feature points, the paper removes the wrong match points by using symmetrical method.An object tracking system is designed and implemented in this paper. The system uses CM-T3730, which is produced by Israel Compu Lab company, as the embedded core board. The core board uses USB camera and IP camera to capture images, realizes the object trackingand also compresses and transfers videos. Since The CM-T3730 is ARM+DSP dual-core architecture, so there are two solutions for video compression. One solution is using the FFmpeg open source libraries to compress the videos. Another solution compresses the videos by using DSP hardware acceleration. The transmission of this system is based on live555 streaming media server, and transfers the videos to the ground station. In the mean time, experiments are performed to verify the improved object tracking method on the designed system. The experiment result is presented to verify the feasibility of the object tracking system designed by this paper. It can be used in various occasions, such as UAV, robotics and vehicle system. |
