Abstract: With the increasing demand of image processing algorithm, a single vision algorithm is often difficult to meet the task requirements. To solve this problem, a multi task processing algorithm with detection, instance segmentation and multi-object tracking is proposed. The Anchor Free framework was used to realize one-stage detection, segmentation and tracking. The grid-based prediction strategy was used to reduce the computational load caused by multi task branching. The tracking branch encoded the tracking objects by word embedding, and associated them according to the distance between the word. The segmentation branch adopted the combination of mask coefficient and original mask to balance the accuracy and speed of the algorithm. The experimental results show that this algorithm can meet the accuracy requirements on the basis of real-time operation.