Feature Point Extraction and Tracking Based on a Local Adaptive Threshold

Abstract

Navigation, environment perception and localization are important capabilities of intelligent vehicles. In this paper, environmental perception and localization from binocular vision are studied. First, an outdoor feature point extraction algorithm that uses a local adaptive threshold is proposed to acquire environmental information. The algorithm filters feature points by setting adaptive parameters and calculating each pixel threshold with a dynamic local threshold. Second, an accurate method for feature point tracking is proposed for localization. We present exhaustive evaluation in 4 major scenarios from the most popular datasets. Evaluating the proposed method with traditional and state-of-the-art extraction methods and experimental results demonstrates that when the brightness decreases or increases, the performance of the proposed method is stable in terms of the number of feature points, the calculation speed and the overall repetition rate. Our proposed tracking method outperforms state-of-the-art tracking methods in terms of the root mean square error (RMSE) and the errors in the dimensions in the scenarios.

Keywords

• Feature extraction
• tracking
• oriented FAST and rotated BRIEF(ORB)
• features from accelerated segment test (FAST)
• root mean squared error (RMSE)