Environmental sensing is a key technology for the development of unmanned cars, drones and robots. Many vision sensors cannot work normally in an environment with insufficient light, and the cost of using multiline LiDAR is relatively high. In this paper, a novel and inexpensive visual navigation sensor based on structured-light vision is proposed for environment sensing. The main research contents of this project include: First, we propose a laser-stripe-detection neural network (LSDNN) that can eliminate the interference of reflective noise and haze noise and realize the highly robust extraction of laser stripes region. Then we use a gray-gravity approach to extract the center of laser stripe and used structured-light model to reconstruct the point clouds of laser center. Then, we design a single-line structured-light sensor, select the optimal parameters for it and build a car-platform for experimental evaluation. This approach was shown to be effective in our experiments and the experimental results show that this method is more accurate and robust in complex environment.
Keywords: laser stripe extraction; semantic segmentation; structured-light vision sensor.