Research On Multi-feature Fusion Detection Based On Support Vector Machine

Pedestrian detection technology has a wide range of applications,providing Pedestrian detection technology provides important technical support for video surveillance,vehicle assisted driving,intelligent robots and other applications,and is a very important branch in the field of computer vision.In real life,there are many problems to be solved in pedestrian detection,such as the variability of clothing,the diversity of posture changes,the complexity of the environment,the difference of light intensity and the mutual occlusion among pedestrians.How to select a feature that is not affected by various external factors and how to train a classifier with strong resolution and robustness is the focus and difficulty in the field of pedestrian detection.Traditional pedestrian detection technology has disadvantages such as long detection time and low accuracy,which has a great impact on subsequent judgment and tracking.Based on the idea of improving detection performance,this paper proposes a multi-feature fusion pedestrian detection method based on SVM.The algorithm used for pedestrian detection was improved.Based on the most classic HOG+SVM detection algorithm,its detection system was completed,performance was tested and data were collected,so as to provide theoretical basis and comparative data for subsequent algorithm improvement.Aiming at the problem that Haar feature detection rate is high but false detection rate is also high,in order to improve the detection accuracy,a double detection method is proposed,that is,the head-shoulder model is first used for detection,and the area of interest marked in one section of detection is then detected in the second section to increase the detection accuracy.In view of the problem of high dimension and large computation caused by feature fusion in two-stage detection,the feature PCA(Principal Component Analysis)dimensionality reduction process was conducted before feature fusion to reduce the original data dimension,so as to maximize the variance between the data with reduced dimension,reduce the required computation,and enhance the timeliness of detection.Experimental comparison shows that the second-order structure detection method has the advantage of Haar feature in detection speed,and the advantage of feature-based fusion algorithm in accuracy.At the same time,secondary detection can effectively solve the problem of excessive error in primary detection.The second-level detection algorithm constructed in this paper can meet the real-time requirements,and the error rate was reduced by about 10 percent compared to the first test.