在智能監(jiān)控領(lǐng)域，實(shí)現(xiàn)人群計(jì)數(shù)具有重要價(jià)值，針對(duì)人群尺度不一、人群密度分布不均及遮擋等問(wèn)題，提出一種多尺度多任務(wù)卷積神經(jīng)網(wǎng)絡(luò)（MMCNN）進(jìn)行人群計(jì)數(shù)的方法。首先提出一種新穎的自適應(yīng)人形核生成密度圖描述人群信息，消除人群遮擋影響；其次通過(guò)構(gòu)建多尺度卷積神經(jīng)網(wǎng)絡(luò)解決人群尺度不一問(wèn)題，以多任務(wù)學(xué)習(xí)機(jī)制同時(shí)估計(jì)密度圖及人群密度等級(jí)，解決人群分布不均問(wèn)題；最后設(shè)計(jì)一種加權(quán)損失函數(shù)，提高人群計(jì)數(shù)準(zhǔn)確率。在UCF_CC_50和World Expo‘10數(shù)據(jù)庫(kù)上進(jìn)行了評(píng)估，驗(yàn)證了自適應(yīng)人形核的有效性。實(shí)驗(yàn)結(jié)果表明：所提算法比Sindagi等的方法 （SINDAGI V A，PATEL V M. CNN-based cascaded multi-task learning of high-level prior and density estimation for crowd counting. Proceedings of the 2017 14th IEEE International Conference on Advanced Video and Signal Based Surveillance. Piscataway，NJ:IEEE，2017:1-6）在UCF_CC_50數(shù)據(jù)庫(kù)上平均絕對(duì)誤差（MAE）數(shù)值和均方誤差（MSE）數(shù)值分別降低約1. 7和45；與Zhang等的方法（ZHANG Y，ZHOU D，CHEN S，et al. Single-image crowd counting via multi-column convolutional neural network. Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern