Multi-object tracking is focused on estimating the location and identity information of objects within a video. Traditional methods often utilize separate detector and reidentification networks for detection, demonstrating high accuracy but inadequate real-time performance. This constraint impedes the practical implementation of multi-object tracking algorithms. Therefore, we introduce an advanced, lightweight detector network that integrates detection and feature embedding. To mitigate the potential accuracy reduction linked with a lightweight network, we introduce a cache appearance rematching association strategy. In conclusion, we present a rapid multi-object tracking methodology that achieves a frame rate of 30.62 on the Multiple Object Tracking 17 dataset while maintaining a commendable performance with 70.50 accuracy and 2277 identity switches. To further enhance its usefulness, we integrate a servo control strategy, deploying the approach on a quadruped robot platform to achieve real-time object tracking in practical settings at a rate of 19.80 frames per second.