In this paper, we treat the parking lot surveillance video as the research object of parking space detection algorithm. Firstly, the algorithm obtained the feature information of the related parking spaces through offline testing in different environment background. Secondly, a single parking area is extracted as the region of interest (ROI). It is used three kinds of parking space detection algorithm to them respectively. The algorithm running on ARM + DSP heterogeneous dualcore processor DM3730 platform and embedded Linux operating system. Both OpenCV image processing function and DSP algorithms for image processing are used that the results can be saved to the embedded database,then more parking space detection based on B/S network model could be realized. Test results show that each user may use the browser to access the system, and get the information about the current parking space occupancy, which can satisfy the requirement of high realtime performance and high accuracy.