Depth Spatiotemporal Information Fusion Moving Target Tracking Algorithm Based on CUDA Library

Mobile object tracking is a computer vision technology that detects, recognizes, and tracks moving objects in video sequences. It is widely used in various fields, such as intelligent security, traffic safety, unmanned aerial vehicle reconnaissance, sports events broadcasting, and wildlife protection. Although mobile object tracking technology has become increasingly mature in practical applications, many performance defects still exist under complex conditions. For example, targets can be occluded by other objects, or may temporarily disappear during motion, increasing the difficulty of association. Occlusion is, therefore, one of the primary challenges faced by current mobile object tracking technology. To address this problem, this paper proposes a CUDA-based motion object tracking algorithm called Deep Space Information Fusion (DSIF), which combines the cluster difference threshold method with template matching to improve the accuracy of mobile object detection. After detecting motion events using the difference threshold method, the method divides events into clusters and uses distance mean and matching templates to redefine motion events to filter noise and interference elements in order to improve the detection accuracy. The algorithm tracks objects based on feature information and temporal-spatial information changes between frames. By using the CUDA library, this algorithm significantly improves computational speed, making it possible to apply it to embedded devices. Experimental results show that the algorithm can lock onto and track multiple different objects in real time.