4.1.1. Pallet Monitoring Module

The monitoring system consists of 8 network RGB cameras and a dispatch server, and 8 network cameras are connected to the server through a switch. We placed 53 pallets in the field of view of camera 8-th and multiple pallets in different locations and numbers in the field of view of other cameras, while placing goods on the pallets to simulate real scenarios. We expect the algorithm to have a good recognition rate for this complex and variable situation. Figure 5 shows the working situation of the monitoring system.

(a) Position of surveillance camera (b) Installation angle of camera

**Figure 5.** Pallet monitoring deployment. The surveillance cameras were mounted as high as possible and angled so that more pallets could be monitored. The network camera was installed at a height of approximately 3.27 m and the network camera was overhead (with vertical lines) at an angle of approximately 50 deg.

#### 4.1.2. Pallet Positioning Module

A RGB-D camera was used for pallet position recognition, which was installed in a fixed position on the forklift and the data were uploaded to the forklift mounted IPC for position calculation via a network cable. In order to avoid the fork tines of the forklift from appearing in the image, we adjusted the camera to a suitable position so that it could see the complete pallet hole position without seeing the fork tines of the forklift. Figure 6 shows the RGB-D camera and its recognition results.

(a) Position of surveillance camera (b) Installation angle of camera

**Figure 6.** Pallet positioning module deployment. An on-board RGB-D camera was used and the camera was adjusted to a suitable position. The 8 corner points of the pallet aperture are recognized and the result of the pose calculation is shown in the upper left corner of the image.
