Neural Network Based Online Feature Selection for Vehicle Tracking

Abstract. Aiming at vehicle tracking with a single moving camera for autonomous driving, this paper presents a strategy of online feature selection combined with related process framework. Detected vehicle can provide more information for tracking. A principal component analysis neural network is used to select appearance features online. Then the positive and negative histogram models using selected features are found for the detected vehicle and the surroundings. A likelihood function is defined based on histogram models, and it can be used as a simple classifier. For selected multiple features, the corresponding multiple classifiers are combined with a single layer perceptron. Experimental results indicate the validity and real-time performance.

• Type: Conference
• Year: 2005
• Where: ISNN
• Authors: Tie Liu, Nanning Zheng, Hong Cheng
• Keywords: Histogram Models, ISNN 2005, Negative Histogram Models, Single Moving Camera