Research On Feature Extraction Method Of Bus Automatic Station System Based On Machine Vision

In this thesis the optimized machine vision and video image processing technology is applied to the bus automatic station reporting system,so as to realize the information and intelligence of the bus station system. Through the camera capture the video along the bus route in real-time and extract key frames of the video, establish image database which save the pictures before the bus station, match the key frames and images in the database. After a successful match then drive the bus-stop system to report the station. This thesis mainly studies the method of image feature extraction, which is a key part of the visual reporting system, and directly affects the real-time and stability of the system.Firstly, experiment shows that the SIFT and SURF algorithms exhibit good performance under a variety of image transformations, such as scale, brightness, or noise, etc. However, we must pay high computational cost and large storage to construct the descriptor, so it could not be used in real-time bus station system. And the greatest strength of the FAST corner is the higher computing speed. A fast robust FAST corner operator is proposed based on the FAST, combined with the histogram equalization algorithm and the Bilateral filter algorithm. Experiments show that the improved-FAST algorithm has higher repetition rate, better illumination invariance and noise immunity.Secondly, a local descriptor LATCH based on binary bits is presented and implemented in this paper, and SIFT, SURF and traditional binary operator BRIEF, ORB is compared. It is proved that the LACTH has the characteristics of high precision and high speed, and its storage requirements is low, so it could be used in real-time application. Based on improve FAST feature points, a feature description method improved FAST+LATCH is presented in this thesis, witch has good real-time performance and high robustness, can be used in application of station system. In addition, the RANSAC algorithm is adopted to eliminate the mismatched point pair, which further improves the matching accuracy.Finally, Use Visual Studio 2015+ Open CV platform to make simulating and experiment for Bus Station, extract key frames of bus video, and establish image database. we compare improved FAST+LATCH algorithm of this thesis with SIFT and SURF, the results show that the improved algorithm has the advantage of high success rate, precision and real-time, which can effectively improve the stability of the system.