Research And Application Of Target Detection And Path Planning Algorithms For Mobile Robots

Mobile robots are booming in today’s technology and have become a popular research direction in various industries and fields.In the practical application of mobile robots,path planning should take into account the possible constraints such as traffic signs,when target detection technology is needed to assist mobile robots in path planning,so target detection and path planning are two key technologies to enhance the autonomous navigation capability of mobile robots.With the continuous development of computer vision,the accuracy of deep learningbased target detection models has significantly improved.However,a concomitant problem is that these models have become bulky,making them difficult to deploy on edge computing devices such as mobile robots,and slowing down detection.Although the proposed lightweight target detection models alleviate these problems to some extent(e.g.,YOLOv4-tiny),lightweight models tend to sacrifice in terms of detection accuracy due to the reduced model parameters.Therefore,this thesis provides an in-depth study and optimization of the YOLOv4-tiny algorithm.A diverse and well-covered traffic sign dataset is constructed to ensure the generalization capability of the algorithm in various scenarios.The CBAM attention mechanism is introduced,optimized with K-means++ clustering and CIo U loss function optimization,allowing the improved YOLOv4-tiny algorithm to maintain a high speed while enhancing the algorithm’s ability to focus on and distinguish between targets,further improving the detection accuracy and localization accuracy of the algorithm.Traditional path planning algorithms are easy to deploy on robotic edge devices,but there are some problems in some aspects,such as path redundancy,more turning points,large turning angles,and path and obstacle collision safety.These problems limit the efficiency and safety of robot movement in real environments.In this thesis,the A*algorithm and the DWA algorithm are improved and fused.The global path planning is improved by weighing heuristic functions,introducing safety coefficients,and using a bidirectional redundant node deletion strategy and cubic B spline curve smoothing;the DWA algorithm is optimized by improving the trajectory evaluation function.In addition,this paper integrates the traffic sign detection algorithm with the path planning algorithm,solves the problem of traffic signs changing the path direction,and further improves the navigation capability of the mobile robot.Finally,in order to verify the feasibility and practicality of the improved algorithm in this paper on the embedded platform,this paper adapts the algorithm to the features of NVIDIA Jetson TX2 platform,builds a ROS mobile robot experimental platform,and conducts a series of experiments.By applying the improved algorithm in real scenarios and evaluating the performance,it is verified that the algorithm not only improves the effect of target detection and path planning,but also has a positive impact on the feasibility and practicality in the field of mobile robotics to promote the frontier technology development in the field of mobile robotics.