Design of intelligent video analysis system based on DM642 platform

With the rapid development of security monitoring systems, new intelligent video analysis technologies are becoming more and more mature. Intelligent video analysis systems are based on image processing technology. In addition to all the functions of traditional security monitoring systems, they also have functions such as threat target detection, identification, tracking and early warning, Ethernet video transmission, etc., which can achieve 24-hour uninterrupted monitoring of complex scenes and automatic early warning.

The currently widely used security monitoring system is based on the traditional PC platform, which is costly, bulky, complex to operate, and has a limited scope of use. This paper uses Texas Instruments (TI) DM642 as the core to design and implement a low-cost intelligent video analysis system. The system analyzes the video stream to detect, identify, track, and warn of threat targets, and realizes video compression transmission through Ethernet. The product can be widely used in the construction of traditional video surveillance projects to improve the intelligence of security monitoring systems.

1 Hardware Composition

This paper uses the DM642 processor from TI of the United States. The core of this processor is the TMS320C64xDSP core with a main frequency of 600 MHz, which can effectively implement complex video processing and analysis algorithms.

The hardware platform of the intelligent video analysis system provides 1 analog video input, 2 RS 232 and 1 Ethernet port, which can be connected to a standard CMOS camera and a monitoring backend. The hardware of the system adopts a modular design, consisting of an image processing board, a power module and a video IP module, as shown in Figure 1.

The image processing board is used to receive the analog video from the camera, analyze and process the video stream, and transmit the processed analog video stream to the video server IP module.

The video server IP module receives the analyzed and processed analog video stream and outputs the encoded and compressed video information to the monitoring backend. It receives the parameter setting and control information from the monitoring backend, transmits the parameter setting information to the image processing board, and transmits the control information to the external on-site alarm device to control the start and stop of the device.

2 Software Architecture and Process

The intelligent video analysis system software mainly includes video processing module, intelligent analysis module, decision module, image compression module, etc. The system software workflow is shown in Figure 2. The interfaces of all algorithms conform to the xDAIS standard of TI.

The video processing module includes data acquisition, data processing and preprocessing, as shown in Figure 3. The analog video signal of the CMOS camera is collected by the A/D chip of the image processing board, and then encoded and output as a standard YUV digital video stream. The image processing board performs video preprocessing on the YUV digital video stream.

The algorithm of the video preprocessing module includes two parts: data processing and preprocessing. Data processing is used to adjust, compress and store the collected video stream for transmission via Ethernet. Preprocessing mainly includes camera calibration, as well as image filtering, enhancement and restoration.

The original code stream captured by the image processing board through the A/D chip has low original image quality due to factors such as illumination, noise, jitter, lens distortion, etc., so it needs to be preprocessed to extract the required information. Before using the camera equipment, the camera lens parameters need to be calibrated, including obtaining the external and internal parameters of the camera lens.

External parameters refer to the position and direction of the camera lens relative to the geodetic coordinate system; internal parameters are the optical characteristic parameters of the camera lens, including the focal length value, radial lens distortion value, axial lens distortion value and other system error parameter values.

The intelligent analysis module mainly detects and measures the target information of interest in the preprocessed image to obtain the objective information of these targets. The intelligent analysis module includes algorithms such as target segmentation, target positioning, target tracking and feature extraction, as shown in Figure 4.

The intelligent decision-making module uses the results of the intelligent analysis module to study the properties of multiple targets in the image and the connections between them, and to understand the meaning of the image content, including the interpretation of continuous scenes, so as to achieve the identification, tracking and early warning of threat targets. The intelligent decision-making module includes algorithms such as target recognition, behavior understanding, threat estimation and decision reasoning, as shown in Figure 5.

3 Conclusion

This design implements an intelligent video analysis system on the DM642 platform. By optimizing the algorithm, it can meet the requirements of 24-hour uninterrupted and efficient monitoring of complex scenes. It has the advantage of low false alarm rate and realizes low-cost standard-definition intelligent video analysis, and has broad application prospects.