Light Field Imaging Distance Measurement Method Based on BP Neural Network

As light field camera can concurrently record light space and angle information, it is possible to use a single image to measure the distance of objects. It is of great significance to three-dimensional reconstruction, image guidance and other practical application. In this article, we analyze the imaging principle of plenoptic 1.0 camera to infer the refocus formula containing distance coefficient. In the unceasing refocusing process, ,the imaging quality of the same object continually varies with the change of refocusing coefficient. For acquiring the refocusing coefficient value of object at the clearest moment, we put forward a BP neural network-based focusing assessment model to quantify image quality, and adopt TID2013 data set to train the model in combination with tester grade . The imaging distance can be calculated according to refocusing coefficient. After bringing the imaging distance into Gaussian imaging formula, we derive the actual distance of the object. Moreover, we respectively test the algorithm and analyze error with simulated light field data and real light field data. The final outcomes suggest that the algorithm shows good distance measurement ability and edge recognition ability in a specific scope.