用于盲人视觉辅助的多目标快速识别并同步测距方法

A fast recognition and synchronously ranging method for multiple targets for blind visual aids applications

提出一种以摄像头实现的可用于盲人视觉辅助的多运动目标快速识别并同步测距方法.该方法以深度学习多目标检测算法(single shot multibox detector,SSD)识别各类目标,并通过SSD输出的目标类别及检测框(bounding box)高度来提出测距模型,从而同步地获取多个目标的测量距离.本方法仅通过普通摄像头便能识别较多类物体且识别类别数量可拓展,能够将测距模块和障碍物识别模块同步执行,从而可对多个物体实时识别并同步测距.实验结果表明,本方法能有效地识别障碍物,具有良好的测距精度,为盲人视觉辅助的一种有效探索.

This work proposes a camera-based method for fast recognition and synchronously ranging of multiple moving targets for blind visual aids applications. This method can quickly recognize a specific target by using multi-objective detection algorithm single shot multibox detector（ SSD） based on deep-learning techniques. Based on the categories and bounding box information obtained via SSD model,a ranging model is proposed to estimate the distance of targets. This method can identify many kinds of objects only by ordinary cameras,and simultaneously identify and range the obstacles. Experiment results show that this method can identify obstacles effectively,estimate the range correctly,and therefore are promising to be applied in practice.