| abstract |
The invention relates to a multi-stage lightweight crowd counting method based on deep learning, comprising the following steps: A. constructing a training data set according to the collected crowd video surveillance data; B. constructing a crowd counting model and setting model training parameters; C. , train the crowd counting model; D. Calculate the evaluation index of the crowd counting model during the training process, if the evaluation index is less than the set value, save the model parameters and execute step E, otherwise execute step C to continue training; E, obtain the same scene The crowd video of the video is predicted by the model every frame of the video, the number of people in the scene is counted in real time, and the crowd density map is displayed. Compared with the prior art, the present invention utilizes the combination of depthwise separable convolution and multi-scale ideas, and improves the counting efficiency of the model under the premise of ensuring the crowd counting effect, making it more practical. |