Algorithm of Detection of Articles Left behind in Vehicles

(Purpose) In order to improve the timeliness and stability of detection of targets in a vehicle and to reduce the probability of loss of articles, (Method) the author uses PyTorch to build a full convolutional neural network model for target detection which adopts ResNet as the backbone network and FPN for extracting the feature maps of higher-order and lower-order network. (Result) After convergence of model, the comparison of the result received from validation set with the effect of target detection in original ResNet network suggests that the feature extraction capacity and stability are improved significantly with this model. (Conclusion) The improved network structure has a good application prospect in detection of targets in a vehicle.