目前相關的行人檢測追蹤辨識系統，因為精準度要求越來越高，就有可能會需要使用到目前較流行的CNN深度學習演算法提高精準度，但相對的對硬體的需求也必須要提高，才不會造成延遲的情況發生，所以一般都建議要搭配一定等級的GPU(Graphics Processing Unit)裝置，才較能達到即時性的速度，此情況可能導致在資源沒那麼充裕的地方，因無法添購GPU裝置，導致較進階的行人檢測追蹤安全系統設置困難，進而影響社會安全之類的提升。故本論文提出結合CamShift和DeepSort的整合式行人追蹤辨識系統，此系統既無需使用GPU裝置加速，也保有一定的追蹤及時性，以CamShift搭配DeepSort的卡爾曼濾波和適時執行Re-ID追蹤行人，此結合CamShift及DeepSort的整合式行人追蹤辨識系統，可因適時調整使用CNN深度學習演算法，增加無GPU裝置的追蹤即時性。實驗結果顯示本研究提出的方法，可以在只使用CPU(Central Processing Unit)的情況下，也能達到不錯的行人追蹤即時性，證明可以使用較少硬體資源也可以有較佳的追蹤辨識效率。;At present, the relevant pedestrian detection, tracking and identification systems may need to use the more popular CNN deep learning algorithm to improve the accuracy because of the increasingly high accuracy requirements, but the relative demand for hardware must also be improved. , it will not cause delay, so it is generally recommended to match a certain level of GPU(Graphics Processing Unit) device to achieve instant speed. The purchase of additional GPU devices makes it difficult to set up a more advanced pedestrian detection and tracking security system, which in turn affects the improvement of social security. Therefore, this paper proposes an integrated pedestrian tracking and identification system combining CamShift and DeepSort. This system does not need to use GPU device acceleration, but also maintains a certain tracking timeliness. CamShift is combined with DeepSort′s Kalman filter and timely execution of Re-ID to track pedestrians. This integrated pedestrian tracking and identification system combined with CamShift and DeepSort can adjust the use of CNN deep learning algorithm in a timely manner to increase the real-time tracking of devices without GPU. The experimental results show that the method proposed in this study can achieve a good real-time pedestrian tracking under the condition of only using the CPU(Central Processing Unit), which proves that it can use less hardware resources and have better tracking and identification efficiency.
