Car depth estimation within a monocular image using a light CNN

Mobile intelligent systems that need to perceive the environment and move in it must measure its depth. Therefore, this issue is pervasive in intelligent devices, especially self-driving cars. Self-driving cars estimate the depth of the surrounding environment and objects using a variety of sensors. However, given the sensitivity of this task, there should be several backup depth estimation systems to minimize the possibility of error. So, it is helpful to design a system that can estimate the depth with higher accuracy and lower computational cost. This paper uses a method to estimate cars' depth within a monocular image. For this purpose, first, a light CNN(MTCNN) detects the license plates of vehicles. Then an MLP neural network that learned the nonlinear perspective relation between license plate dimensions and its depth estimates the depth of the cars within the image according to the coordinates and dimensions of the license plate bounding box.