Lightweight gesture real-time detection method based on improved SSD

The invention discloses a lightweight gesture real-time detection method based on an improved SSD (Solid State Disk), and the detection method takes the SSD as a basic network, and comprises the steps: taking Mobile Net v2 as a model trunk feature extraction network, and reducing the parameter quantity and calculation complexity of a model; designing an INA multi-scale convolution module, applying the INA multi-scale convolution module to three prediction feature layers, and increasing the adaptability of the network to different scale features by connecting convolution kernels of different sizes; a K-means + + clustering algorithm is adopted to adaptively generate a candidate frame suitable for the hand, and the hand is accurately positioned to improve the detection precision of the model; and training the improved SSD network structure on the manufactured gesture data set to obtain a trained gesture detection model for gesture detection. By applying the method, the problem that a hand detection model is dif