Support Vector Machine (svm) In The Application Of The Pedestrian Detection

Intelligent Transportation Systems (IntelligentTransportationSystem, referred to as ITS) is the future direction of development of the transport system. Pedestrian detection technology is one of the key technologies of intelligent transportation systems, The goal is to quickly and accurately judge the scenes in the captured image from the camera if the location of the pedestrian as well as the positioning of pedestrians for the system to alert or alarm to the driver, to avoid the collision of vehicles and pedestrians.Support vector machines in recent years, extensive attention and the development of a statistical learning algorithm, it boils down to solving a convex quadratic programming problem. SMO algorithm for support vector machines specially designed algorithm, the advantage is that each iteration only the optimization of the two multipliers, so you can get the analytical solution, to avoid the numerical algorithm. Select the optimal point is heuristic, and thus the convergence speed of magnitude increase.In this paper, the standard SMO algorithm proposed an improved SMO algorithm to select the second optimization multiply neutrons have three options, the first program can cause the growth point of the objective function, if this point optimization fails to adopt the secondprogram, that is, a traversal of all non-boundary samples, until you find an optimal point, should this also fails the third program from all samples to find points that can be optimized. Improvements of this article is for the second program, order traversal from a random location, but in accordance with the increase of the objective function can cause a rough sort of (strict sorting would greatly increase the space and time complexity of the algorithm, but reduce the performance of the algorithm), in this order to try to optimize until it succeeds. In this way, each time optimization multiply neutrons can always cause a larger objective function growth, despite an increase in the work done in each optimization, but never a whole to reduce the number of optimized to shorten the running time. This improvement is the possibility of large enhancement to the performance of the algorithm depends how likely will fail in the first program, the experiments show that:in the data set used in this article, the first program in the first traversal of the entire sample set success rate is greatly reduced when a high success rate, but in the second even after traversing the entire sample set, while the second program, a high success rate, therefore the improved algorithm able to play a role, from the experimental results, the performance of the algorithm achieved gratifying upgrade.