|
Capturing motion vehicle information from satellite videos is crucial for real-time traffic monitoring and emergency response. However, vehicles in satellite videos are small in size, lack detailed textural features and are easily obscured by complex backgrounds. Traditional frame differencing and background subtraction methods often lead to a high number of false positives and false negatives, while deep learning methods struggle to meet real-time processing requirements. To strike a balance between detection performance and processing efficiency, a real-time vehicle detection method combining hole features extracted from frame differencing and motion direction estimation is proposed. Initially, a multiframe image accumulation (MIA) strategy is employed to enhance the visibility of vehicle targets and suppress background noise. Subsequently, the hole features of moving vehicles are extracted by differencing adjacent accumulated images, leading to the creation of a hole model for coarse detection of moving vehicles. Then, by estimating the motion direction of vehicles and enforcing direction consistency constraints, hole matching of neighboring vehicles in motion is achieved to improve the detection accuracy. Finally, a novel region extraction algorithm that integrates target hole features and motion direction information is designed to effectively suppress false positives generated by background noise. This method exhibits superior detection performance on the VISO benchmark dataset, achieving a recognition accuracy of 89.5% while meeting real-time processing requirements with an average processing time of only 0.04 seconds per frame, ensuring both detection performance and processing efficiency.
|
