Analysis of the Effect of Automotive Ethernet Camera Image Quality on Object Detection Models
Huang Hsiang, Kuan‐Chung Chen, Po-Yi Li, Yung‐Yuan Chen · 2020
This paper centers on analyzing the effect of automotive Ethernet camera image quality on the object detection models. We choose the five pre-trained models from the TensorFlow model zoo and set various image bitrate to the comparison of detection accuracy on different vehicle object sizes. We use an automotive Ethernet camera to collect image dataset on actual roads and different weather. The experiments were performed to obtain the detection accuracy of each detection model for different object sizes under different image quality and hardware platforms. The contribution of this study is to show how to choose a feasible object detection model in terms of the bandwidth, image quality, object size, cost, detection speed, and detection accuracy requirements.