Non-line-of-sight fast tracking in a corridor

Abstract

Non-line-of-sight (NLOS) detecting technology is used to locate and track a hidden object through scattered light after three reflections. This technology is expected to be applied in automatic driving, danger rescue, etc. Herein, we improve the real-time performance of NLOS detection in two aspects. Firstly, we introduce the method of dynamic programming to detect extremely weak signals from the hidden object based on current and previous information, which can unmix the target signal from noise at a shorter exposure time. Secondly, we simplify the positioning method by extracting the time difference from the histogram and training a feed-forward network that speeds up the process of spatial positioning. Based on the improvements, we demonstrated fast tracking experiments in a corridor 20 m away, tracking a moving hidden object at framerates of 50 Hz. These improvements make NLOS fast detection a step forward.