Object Detection Method of Autonomous Vehicle Based on Lightweight Deep Learning

Object detection is an important visual content of the autonomous vehicle, the traditional detecting methods usually cost a lot of computational memory and elapsed time. This paper proposes to use lightweight deep convolutional neural network (MobilenetV3-SSDLite) to carry out the object detection task of autonomous vehicles. Simulation analysis based on this method is implemented, the feature layer obtained after h-swish activation function in the first Conv of the 13th bottleneck module in MobilenetV3 is taken as the first effective feature layer, and the feature layer before pooling and convolution of the antepenultimate layer in MobilenetV3 is taken as the second effective feature layer, and these two feature layers are extracted from the MobilenetV3 network. The two output effective feature layers of MobilenetV3 are used to replace the two output feature layers of VGG16 in the SSD network, and the rest convolutions of SSD was replaced by lightweight separable convolutions and inverted residual structures, thus the architecture of Mobilenetv3-SSDLite is completed. It trained 800 epochs on RTX 2080 using KITTI data, and its training and detecting performance is recorded. In order to be evaluated, this method is compared with the original SSD networks from training and detecting performance. Results show that both MobilenetV3-SSDLite and original SSD networks can detect the pedestrians and vehicles effectively, but MobilenetV3-SSDLite network can greatly improve the detecting speed and save the hardware cost by only slightly sacrificing the detecting accuracy, therefore, it is more suitable for mobile platform such as the autonomous vehicle, and then it is more suitable for object detection task of the autonomous vehicle.