Quantization Algorithm for Human Foot Detection Based on Convolutional Neural Networks

Abstract:
With the rapid development of the artificial intelligence industry, the level of automobile intelligence is getting higher and higher. Among them, the inductive electric tailgate system, as an important configuration for high-end cars to enhance user experience, also uses various sensing technologies. The traditional inductive electric tailgate uses a capacitive electric field sensor to trigger the electric tailgate switch by analyzing the electric field signal changes of two electrodes installed at different positions. The biggest drawback of this method is that it is easy to touch by mistake. Once an object is swept under the tailgate, the electric switch will be triggered to open the trunk. To this end, based on the NanoDet network model, this paper proposes a quantification algorithm for human foot detection based on a convolutional neural network to detect whether the position of the user's foot is within a preset range, to determine the user's intention to open the door. Through the deep learning target detection method, the target position in the natural scene is identified, the model is compressed and quantized, and the floating-point operation is converted into a fixed-point operation, which can not only quickly detect the position of the human foot, but also facilitate the deployment of the model to the edge computing platform. In this paper, we choose the PyTorch deep learning framework and validate the proposed algorithm in Ubuntu 18.04 operating system. The experimental results show that the quantization algorithm compresses the network model from 7.1MB to 1.1MB, the compression ratio reaches 84.5 %, and the accuracy rate only drops by 1.2%. The experimental data illustrate that after quantizing the parameters of the convolutional neural network, the algorithm can effectively compress the model with a small loss of accuracy, which solves the problem that the convolutional neural network is difficult to deploy to the edge computing platform.