Long-Run Real-Time Human Tracking at Low Light Condition

In this paper, we present a novel algorithm Long-Run Low-Light Tracker (LL_Tracker) for long-run, real-time object tracking in challenging environmental conditions such as low light and occlusion. The proposed LL_Tracker algorithm utilizes Retinex based color-enhancement along with Kalman filter and template matching-based object identification and tracking. Furthermore, the algorithm follows a multi-objective optimization approach to improve tracking accuracy. We compare the performance of LL_Tracker algorithm to several existing tracking algorithms including KCF, CSRT, GOTURN, MIL, and MOSSE in terms of tracking duration, accuracy, and robustness under occlusion and low light conditions. The results show that the LL_Tracker algorithm outperforms all other algorithms in each evaluation parameter. It tracks accurately for 100% of the video length achieving superior performance in challenging environmental conditions and real-time compatibility while others stop at 50% or below of their duration. A demo of the system and the source code can be found on https://github.com/Siva3551/LLT-for-long-run-low-light-tracking