A Rapid Multi-Angle Vehicle Detection Method in Complex Scenario

Abstract

Vehicle detection algorithm based on video analysis is negative for vehicle detection in complex scenes, such as traffic intersection, due to the single angle and the poor static detection effect, and so on. Aimed at above problems, this article proposes a multi-angle vehicle detection method, based on micro cascading neural network. This method improves the tertiary AdaBoost recursive model and constructs an embedded micro neural network model. This paper combining the model with local normalized pixel-value differencing (NPD) features, trained AdaBoost model by polymorphic complex angle samples. The results show that the method has better detection performance and lower failure rate, and average detection rate and detection time are 89.47% and 199 ms respectively, which can meet the requirements of real-time vehicle detection in actual scene.