A low-cost learning platform for SLAM, Autonomous Driving, and Indoor Racing
The car body is from the Waveshare Jetracer AI Kit (link). You can 3d print the frame and the alternative camera mounting solution found in the assets folder. The .stp and the original .f3z file is also in this folder should you wish to modify the designs.
The schmematic of the board provided in the kit differs slightly in the motor driver section. Below is the revised schematic that was created by manually probing the connections. It implements two DRV8870 PWM controlled H-bridge ICs to control the left and right motors independently.
After converting the camera image to HSV space, thresholding the output and applying a warp function (warp points shown top right), we obtained the second image. Summing the number of white pixels in the vertical axis along the frameWidth gave a histogram (shown bottom left), which allowed the initialization of the slidingwindow algorithm. The mid-points of our windows were used to estimate the lanes via a polyfit approach and the detected lane is shown on the bottom right.
Due to the narrow angle view of the Raspberry camera, there arises some occasions where only one of the lanes is detected in the warp view. The demo below was applied to images obtained from manual control of the vehicle with a wireless joystick. One goal for the autonomous driving controller is that both lanes will remain in frame, despite the vehicle driving at high speeds.
