Scene-Adaptive Vehicle Detection Algorithm Based on a Composite Deep Structure

Existing machine-learning-based vehicle detection algorithms for intelligent vehicles have an obvious disadvantage in that the detection effect decreases dramatically when the distribution of training samples and the scene target samples do not match. To address this issue, a scene-adaptive vehicle detection algorithm based on a composite deep structure is proposed in this paper. Inspired by the Bagging (Bootstrap aggregating) mechanism, multiple relatively independent source samples are first used to build multiple classifiers and then voting is used to generate target training samples with confidence scores. The automatic feature extraction ability of deep convolutional neural network is then used to perform source-target scene feature similarity calculations with a deep auto-encoder in order to design a composite deep-structure-based scene-adaptive classifier and its training method. Experiments on the KITTI data set and a data set captured by our group demonstrate that the proposed method performs better than existing machine-learning-based vehicle detection methods. In addition, compared with existing scene-adaptive object detection methods, our method improves the detection rate by an average of approximately 3%.