Preceding vehicle image recognition based on multivariate feature information fusion(PDF)
长安大学学报（自然科学版）[ISSN:1006-6977/CN:61-1281/TN]
- Issue:
- 2016年04期
- Page:
- 79-85
- Research Field:
- 交通工程
- Publishing date:
Info
- Title:
- Preceding vehicle image recognition based on multivariate feature information fusion
- Author(s):
- YANG Wei; GONG Jian-qiang; WEI Lang
- (1. School of Automobile, Chang’an University, Xi’an 710064, Shaanxi, China; 2. Research Institute of Highway, Ministry of Transport, Beijing 100088, China)
- Keywords:
- traffic engineering; preceding vehicle; multivariate feature information; Gabor filter; AdaBoost classifier; machine learning
- PACS:
- U491.116; TP391
- DOI:
- -
- Abstract:
- In order to reduce the risk of a false target vehicle detection, this paper proposed a preceding vehicle image recognition method based on multivariate features information matching. It first detected the candidate region of preceding vehicle by analyzing the gray average mutation characteristics, and then used dual-channel Gabor filter to extract vehicle sample picture multiscale directional features. Merging with AdaBoost classifier and Cascade classifier to form a series of strong classifier, the dimension of a high-dimensional feature with 5 scales and 8 directions vector dimension was reduced, meanwhile classifying and extracting feature samples. Finally, according to gray information entropy normalized symmetry, preceding target vehicle precise detection was finished. The results show that the detection accuracy of preceding vehicle is 96.7%, increasing by 1.6% than classical algorithms, the longest whole detection time-consuming is 35 ms, the shortest time-consumingis15 ms, and the average time-consuming is 25 ms.The detection time-consuming is determined by vehicle size and complex degree of background. Besides, the loss of local effective identify information can be avoided under the single feature. It also has better identification precision and processing speed, and the false detection rate is only 3.2%, which is better than the other error detection rate of vehicle recognition algorithm, and thus improves the identification of false target detection.
Last Update: 2016-07-29