In light of the burgeoning advancements in deep learning, object detection and tracking technologies leveraging surveillance camera infrastructure have emerged as a paramount research focus within computer vision. While multi-object tracking algorithms have long been instrumental in security surveillance applications for monitoring and analyzing human behavior, similar requirements have surfaced in tobacco logistics. Concerning target detection, the erstwhile reliance on sliding window methodologies, plagued by prolonged processing times and elevated false positive rates when scanning entire scenes, has given way to deep learning solutions, which offer enhanced efficiency. This study, therefore, specifically addresses the exigency of detecting illicit activities involving tobacco cage trucks during transportation, presenting a novel DeepSORT-based algorithm for concurrent human and cage truck detection and tracking. The proposed framework is designed to proficiently monitor the movement of individuals and vehicles involved, and automatically identify and report any human violations, thereby facilitating the automation and liberation of human labor in this domain. The main research of this paper is as follows, Harnessing the cutting-edge YOLOv7 algorithm for the detection and classification of humans and cage trucks, thereby significantly enhancing the speed and precision of the detection process. Implementing the DeepSORT algorithm to track and annotate humans and cage trucks, generating valuable time-series data that inform subsequent violation detection procedures. Within the context of the violation detection module, the cost matrix is constructed by calculating the Euclidean distance between the detection frames corresponding to the human and the cage-caged vehicle. The LAPJV algorithm is then employed to establish associations between individuals and their respective cage trucks, followed by the detection and recording of any rule infractions committed by individuals whose actions surpass pre-established violation thresholds.