High Resolution Image Reconstruction Of Traffic Surveillance Video Based On DM8127 DSP

In traffic video monitoring system, vehicles and pedestrians need to be tracked and recognized in real time. The resolution of image can be improved withsuper-resolution reconstruction technique, which can provide more accurate information for tracking and recognizing.The research content is originated from Research of traffic video monitoring system Based on IPNCDM8127 in Jiangsu Yitong High-tech Company Limited. In this paper, surveillance video captured by IPNC DM8127 istransferred to serverthrough network, where super-resolution reconstruction is implemented on vehicle images. Thispaper deals with the following researches:Firstly, system platform and developmentenvironment are constructed, based on IPNC DM8127 hardware platform and IPNC RDK software development environment. Meanwhile, video capture, display and network transmissionare realized by programmingbased on Multi-Channel Framework. Furthermore, DSPALG Link is created, which is used to transplantthevehicle images graying and moving vehicle detectionalgorithminto DSP core.Experiments show a good result of moving vehicle detection. Besides, video capture, network transmission and video display are realized in real time which lays a foundation for the transplant of image processing algorithm into DSP.Secondly, aiming at the edge sawtooth and blur of the IBP results, an improved IBP method based on NEDI interpolation is proposed. The factorthat causesedge sawtooth and blur is analyzed. To solve the problem, the proposed method utilizes NEDI interpolation to get the initial estimation of high-resolution image. Experiments are conducted on standard images and real images extracted from traffic surveillance video.Simulation results indicate that the improved method increases peak signal-to-noise ratio and decreases root mean square error. Besides, it improves the edge quality and subjective visual effect of reconstructed image.Finally, as the previous proposed algorithm suffersfrom heavy computation andedge artifacts, a further improved IBP method based on ICBI interpolation is presented. According to the definedenergy term, an iterative refinementis performed on the initial value of interpolated pixels obtained by FCBI interpolationin order to get the final result. The ICBI interpolation which can preserve the edges with low computationis used to get the initial value of high-resolution image.Experiments performed on standard images and real vehicleimages demonstratebetter visual effect,better objective evaluation criteriaand smaller computation time with the further improved method.