Algorithm Implementation of Self-aiming Robot Image Recognition Based on Yolov5s

In view of the problems of low target detection accuracy and poor effect, in order to realize the speed of RoboMaster automatic aiming robot image recognition, this paper proposes an algorithm of automatic aiming robot image recognition based on Yolov5s model. First, the Hikon camera was used to collect the data set for the image recognition of the automatic targeting robot. Second, the labelme was used to annotate the points of the data set. Then, the Yolov5s neural network model was used to train the image recognition of the self-aiming robot, verify and test. Finally, the Yolov5s model is exported in TensorRT format and integrated into the software system of the RoboMaster self-targeting robot, so as to facilitate better deployment and reasoning acceleration. The experiment showed that the self-aiming robot based on Yolov5s model compared the class identification, the inference time increased from 35ms to 15ms, and the memory consumption from 1767M to 1049M; in the combination of traditional vision and neural network, the identification accuracy from 0 m to 3 m distance increased from 90% to 99.2%; from 3 m to 6 m distance, from 78% to 96%. The reasoning speed and object detection ability of the image recognition of the self-aiming robot have been improved. The proposed Yolov5s model of automatic targeting robot image recognition can meet the accuracy and real-time requirements of autonomous target recognition of the self-targeting robot in a complex battlefield environment.