Load forecasting is a key part of system optimization control of air conditioning system, and the randomness and uncertainty of personnel movement make it difficult to accurately estimate the load of building space personnel, resulting in poor control effect of existing control strategies, the slow response of the system, the waste of energy and the reduction of thermal comfort of the building interior environment. To solve these problems, an air conditioning terminal and new air volume classification control strategy based on crowd density estimation is proposed in this paper. Firstly, the image information of building space was collected and a model of multi-column convolution neural network crowd density estimation was established to obtain the number and dynamic distribution of personnel and calculate the real-time load of building space personnel. Secondly, the personnel load control factor was introduced and the air conditioning grading control strategy was put forward to realize the air conditioning terminal and fresh air supply. The experimental results show that the method proposed in this paper can better maintain the internal environment stability of the building, at the same time, the system has faster response speed and greater energy saving potential.