| With the development of video surveillance technology, different kinds of monitoring equipments have been applicated in many places, such as factories, shopping centers, flats, banks and so on. Today, a new requirement is proposed that the video should have a broader view, which means that people could see more contents in a same video scene. To meet this demand, the current common approach is to employ multiple cameras to cover an arear or adopt a speed dome camera to scan back and forth. However, this method can not fully meet the requirements in some cases, when customers need to use the targets for video analysis in a one frame. To solove this problem, a camera that can obtain the panoramic video scene is needed. On the other hand, in order to emancipate the manpower and improve monitoring quality, security field has been committed to smart camera research. Meanwhile, moving object detection is the basis for intelligent video analysis. Therefore, employing panoramic video which has no dead angle to implement multi-object detection can not only expand the monitoring range, but also lay a good foundation for further intelligent video analysis applications.In this dissertation, a multi-object detection software based on the TI DaVinci processor DM6467T is designed for panoramic video. And the processor adopts the ARM+DSP dual-core architecture, Codec Engine framwork and the embedded Linux system. Our main works are as follows:(1) Set up the Linux development environment on DM6467T processor, and design the software framework combining with the software requirements analysis.(2) Panoramic video captured by catadioptric lens does not meet the habits of hunman’s vision. Thus, we design and implement the unwrapping method of panoramic video on the DaVinci DM6467T processor. The speed of optimized unwrapping method can reach 150 per second for Dl resolution.(3) Achieve the multi-object detection of panoramic video on DaVinci processor by using MHOEI algorithm. The experimental results show that the detection method can not only detect the moving object, but also the static object after moving. After optimization, the speed of detection method can reach 80fps. Furthermore, the XDM standard algorithm package of multi-target detection is implemented to facilitate the algorithm calls and transplantation.(4) Design and implement the software function modules, including two channel video capture, video organization and display, two channel video encoding and bit stream upload. |
