Driving scene lightweight human-like target detection method based on multi-task learning

The invention provides a driving scene lightweight human-like target detection method based on multi-task learning. The method comprises the following steps: carrying out real-time environment perception and image acquisition through a visual sensor of an automatic driving automobile; preprocessing the acquired image through a multi-task data processing module to obtain a preprocessed image; constructing a multi-task network model through a multi-task network model module based on the preprocessed image; through a multi-task model training module, continuously optimizing model parameters of the multi-task network model by using a back propagation and gradient descent method until a loss function is converged; updating the weight of the loss function through a multi-task model test module, and screening an optimal solution; and deploying the trained multi-task network model into an intelligent driving system to realize real-time processing of humanoid target detection and attention prediction related tasks. Ac