With the development of transportation intellectualization, forwarding encrypted highway surveillance videos to the cloud has become a major trend in transportation development. Deep transportation data mining, especially pedestrian detection, is one of the crucial problems to be solved in this trend. In this study, to address the problem of pedestrian detection in various road environments, we propose an all-weather pedestrian monitoring solution based on the Kunpeng Cloud. The video streams in the surveillance camera are forwarded to the Kunpeng Cloud through a streaming service. Then, the Kunpeng Cloud decodes the video streams, detects pedestrians, and saves pedestrian history information. Finally, it analyzes and reports the pedestrian events. This system uses an embedded neural-network processing unit (NPU) instead of a traditional graphics processing unit (GPU) platform to accelerate the reasoning of the YOLOv4 pedestrian detection module. The solution not only achieves a fast detection speed and can process 22 video streams in real time but also delivers better results in detecting pedestrians on highways in different road scenes.