Package detection and intelligent sorting

Abstract

To provide a sorting conveyor and a method for sorting packages to selected destinations.SOLUTION: A neural network executed by a multi-core classification processor 36 uses captured images of package units conveyed on an infeed conveyor 18, 19 to classify each package unit. A control processor 34 controls a sorting conveyor 26 to direct each package unit to a destination depending on the package unit's classification.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates generally to powered conveyors, and more particularly to conveyors that use artificial intelligence to classify parcels and sort the parcels to destinations according to their classification. [Background technology]

[0002] In the parcel handling industry, sorting conveyors are used to separate individual parcels from a load stream of randomly oriented, stacked parcels of various sizes and shapes. However, before the parcels can be sorted to their appropriate destinations, they need to be separated from one another. Conveyors use a variety of techniques to separate the load stream into individual parcels. However, occasionally the conveyors fail to separate all of the parcels, requiring manual intervention. Oversized boxes can also be problematic by causing bottlenecks, and when this is recognized, the boxes are manually removed. However, oversized plastic bags may not cause bottlenecks due to their flexibility. Thus, removing these bags from the sorter reduces overall throughput. Summary of the Invention

[0003] One type of conveyor system embodying features of the invention includes an infeed conveyor section that conveys separated parcel units downstream in a conveying direction. A parcel unit detector detects the parcel units at detection locations along the infeed conveyor section. An imaging system captures images of the parcel units within a target zone of the infeed conveyor section. A computer processing system executes program instructions that track the position of each of the parcel units as they are conveyed on the infeed conveyor section and provide each of the images of the parcel units as input to a classifier. The classifier is trained to recognize a set of parcel units and to assign to each of the parcel units a classification that corresponds to one of the members of the set of parcel units that the classifier is trained to recognize. A downstream conveyor section receives the parcel units from the infeed conveyor. The computer processing system executes program instructions that control the downstream conveyor section to convey each parcel unit according to its parcel unit classification.

[0004] Another type of conveyor system includes an infeed conveyor section that conveys parcel units downstream in a conveying direction at a conveying speed. A footprint detector is disposed at a detection location along the infeed conveyor section. A camera is disposed along the infeed conveyor section downstream from the detection location to capture images of a capture area on the infeed conveyor section. A sortation conveyor section receives parcel units from the infeed conveyor section and selectively sorts the parcel units to a plurality of destinations. A control processor executes: (1) program instructions for operating the footprint detector to detect the footprint of parcel units advancing in the conveying direction past the detection location; and (2) program instructions for controlling the sortation conveyor section and the infeed conveyor section. A sorting processor in communication with the control processor executes: (1) program instructions for calculating a position of each of the parcel units on the infeed conveyor section from their footprints detected as they pass through the detection position as they advance along the conveyor; (2) program instructions for controlling a camera to capture an image of one or more of the parcel units if the calculated position of one or more of the parcel units is within a target zone of a capture area on the infeed conveyor section to generate an image of one or more of the parcel units in the capture area; (3) program instructions for cropping the image into one or more cropped images, each cropped image corresponding to one of the one or more parcel units in the capture area; and (4) program instructions for classifying the cropped images into a plurality of classifications using artificial intelligence by assigning one of the classifications to each of the one or more parcel units. The control processor controls the sorting conveyor section to sort each of the one or more parcel units to a destination according to the classification of the parcel unit.

[0005] In another aspect, a processor-implemented method for sorting parcels transported on an infeed conveyor section includes: (a) detecting multiple parcel units advancing on the infeed conveyor section in a conveying direction at a conveying speed past a detection location; (b) calculating the position of each of the multiple parcel units detected as passing the detection location on the infeed conveyor section as they advance along the conveyor; (c) imaging one or more of the multiple parcel units within an image capture area on the infeed conveyor section to generate images of the one or more of the multiple parcel units within the capture area; (d) cropping the images into one or more cropped images, each cropped image corresponding to one of the one or more multiple parcel units in the image capture area; (e) classifying the cropped images into a plurality of classifications using artificial intelligence by assigning one of the classifications to each of the one or more multiple parcel units; and (f) sorting each of the one or more multiple parcel units to a destination according to the classification of the parcel unit. [Brief description of the drawings]

[0006] [Figure 1] 1 is a top view of a transport system embodying features of the present invention; [Diagram 2] 2 is a flow diagram of program instructions executed by the multi-core classification processor and control processor for the conveyor system of FIG. 1. [Diagram 3] 3 is an expanded flow diagram of the TCP connection management task of FIG. 2. [Figure 4] 3 is an expanded flow diagram of the parcel tracking task of FIG. 2. [Diagram 5] FIG. 3 is an expanded flow diagram of the image processing task and the parcel classification task of FIG. 2. [Figure 6] FIG. 2 is a schematic diagram of an exemplary neural network used to help explain the operation of a neural network usable in the conveyor system of FIG. 1. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0007] A conveyor system embodying features of the present invention for intelligently sorting parcels is shown in Figure 1. Conveyor 10 includes an infeed conveyor section 12 which conveys parcels 13 downstream in a conveying direction 14. A parcel unit detector 16 is disposed along infeed conveyor section 12 at a detection location. In this example, infeed conveyor section 12 is illustrated as a series of two conveyors: a first conveyor 18 upstream of parcel unit detector 16 and a second conveyor 19 downstream of the parcel unit detector. Both conveyors 18, 19 may be realized, for example, as belt conveyors, slat conveyors, chain conveyors or roller conveyors. Infeed conveyor section 12 may be realized as a single conveyor or as more than two conveyors. In this two-conveyor configuration, the parcel unit detector 16 emits a curtain of light from the transmitter array through the gap 20 between the two conveyors 18, 19 to the receiver array. The receivers and transmitters are located on either side of the conveyor section 12, one below and one above. The resulting curtain of light spans the width of the conveyor section 12. A parcel 13 passing through the curtain of light interrupts the transmit beam and prevents it from reaching the receiver array. As a parcel 13 passes the parcel unit detector 16, its footprint, i.e., its projection onto the infeed conveyor section, can be determined from the pattern of interrupted receivers. Thus, in this configuration, the parcel unit detector 16 is a footprint detector. Other types of parcel unit detectors, such as laser range finders or cameras, could be used instead.

[0008] Located downstream of the parcel unit detector 16 and along the infeed conveyor section 12 is an imaging system including a camera 22. The camera 22 has a capture area 24 covering a stretch of the conveying side of the infeed conveyor section 12 across its width. The imaging system captures images of the parcels 13 passing through the capture area 24. Parcels 13 leaving the infeed conveyor section 12 are received on a downstream sorting conveyor section 25. The sorting conveyor section 25 may be implemented as a sorting conveyor 26, for example as a roller belt having rollers that are selectively actuated to rotate to divert the articles to one of two flanking conveyors 28, 29. The sorting conveyor 26 and the flanking conveyors 28, 29 may be powered roller conveyors or modular plastic belt conveyors, just two examples. Alternatively, the infeed conveyor section 12 and the sorting conveyor 26 may be implemented by a single conveyor belt. In one form, the sorting conveyor 26 is an INTRALOX® Series 7000 activated-roller-belt conveyor manufactured and sold by Intralox, LLC, Harahan, Louisiana, USA. The flanking conveyors 28, 29 may likewise be implemented by roller belts, flat belts, modular plastic belts, slotted conveyors, chutes, or roller conveyors, for example. The sorting conveyor, by selectively actuating its rollers, diverts parcels to one of three destinations: (1) a right flanking conveyor 28; (2) a left flanking conveyor 29; or (3) a destination 30, which is shown to be staffed by a human worker 32 who can decide how to dispose of the received parcel units.

[0009] The output of the footprint detector 16 is sent to a computer processing system 33 that includes a control processor 34 programmed to determine the footprint of a parcel 13 passing the detector at a detection location. (The computer processing system 33 includes a program memory that stores constants and program instructions executed by the processor or processors, and a volatile data memory that stores calculations, tables, and other temporary or changeable information.) Some parcels may not be separated from other parcels. These overlapping or stacked parcels have a footprint that may be of a different shape than the footprint of a single separated parcel. For this reason, the footprint detector 16 detects the footprint of a separated parcel unit that encompasses a single separated parcel, and the footprint of a group of overlapping or stacked parcels whose footprint is defined by a single contour line.

[0010] The control processor 34 executes stored program instructions which define the individual footprints of parcel units passing the detection location. For example, each footprint may be defined by the coordinates of its corners or by its center of gravity calculated by the control processor 34. The position of each parcel unit on the infeed conveyor can at any time be described by its coordinates in an xy coordinate system which, for example, references the detection location as x=0 and the right edge 35 of the infeed conveyor section 12 as y=0.

[0011] In that arbitrary coordinate system, the x-axis is parallel to the conveying direction 14 and the y-axis is perpendicular to the conveying direction. The control processor 34 can also be programmed to: (a) control the speeds of the infeed conveyor section 12, the sorting conveyor 26, and the first and second conveyors 28, 29; (b) receive input reporting the speeds of the various conveyors; and (c) control a communication network between itself and a classification processor 36 used with the imaging system. Like the control processor 34, the classification processor 36 is contained within the computer processing system 33 and may include an external or same-chip graphics processing unit (GPU). The control processor 34 may interface with the conveyors 12, 26, 28, 29 via an external physical programmable logic controller (PLC) 38, or via an internal virtual PLC, or may share control between an external physical PLC and an internal virtual PLC. For example, an external PLC could control the motors that drive the infeed and sort conveyor sections and could read sensors that report belt speed, while a virtual PLC receives the output of an occupancy detector. In either case, PLC 38 is considered part of control processor 34 and is considered part of computer processing system 33.

[0012] The operation of control processor 34 and classification processor 36 will now be described with reference to the flow diagrams of FIGS.

[0013] The TCP (Transmission Control Protocol) connection manager task 40 executed by control processor 34 (FIG. 1) is shown in more detail in FIG. 3. This task controls a communications network 42 (FIG. 1) that is responsible for communications between control processor 34, classification processor 36, and PLC 38. TCP connection manager task 40 has three subtasks: (1) a read task 44; (2) a connection monitor task 46; and (3) a write event handler task 48.

[0014] Read task 44 reads message 50, parses message 52, and executes message 54. In the event of a message from control processor 34 indicating that a footprint corresponding to a parcel unit has been identified by the footprint detector, an entry corresponding to that footprint and its coordinates is added to the infeed conveyor status table, which represents the most recently calculated position of any identified footprint of the parcel unit on the infeed conveyor 19 downstream of the detection location.

[0015] The write event handler 48 processes any event 56 that occurs, frames a message indicating the occurrence of that event along with any associated data 58 regarding that event, and transmits the message frame 60 over the communications network to be read by the intended recipient. An example is provided below.

[0016] The connectivity monitor task 46 checks to ensure that every device or node, e.g., control processor 34, classification processor 36, and PLC 38, is connected. This task sends out heartbeat messages 62 over the communications network. This task determines 64 whether the message was received by the intended recipient device. If there are no message transmission failures, the control processor knows the network is intact 66. If there are transmission failures, the control processor attempts to reconnect 68 to disconnected devices.

[0017] The sorting processor 36 executes stored program instructions, including a timer-tick task 70 as shown in FIG. 2, at a periodic rate set by a timer, for example every 75 ms. The timer-tick task 70 commands several subtasks, which are tracking parcel units on the infeed conveyor section, capturing images of the parcel units, and classifying the parcel units. First, the timer-tick task 70 clears 72 old subtasks and prepares them for a new execution. The sorting processor 36 executes a parcel tracking task 74 for each parcel unit identification footprint in the infeed conveyor status table. In this form, the sorting processor 36 has multiple cores that can execute the parcel tracking task 74 simultaneously in parallel. The task for each detected parcel unit is executed in one or another of these cores, or in a dedicated thread in one core running multiple threads. By executing the parcel tracking task 74 and subsequent tasks in parallel on separate cores or threads, the conveyor system can accommodate high parcel throughput rates. Similarly, the control processor 34 could potentially be a multi-core processor.

[0018] 4, a parcel tracking task 74 for each parcel unit on the infeed conveyor section calculates 76 the time interval since the previous execution of the task for that parcel unit. Using conveyance speed information, task 74 calculates 78 the distance the parcel unit has advanced in the conveyance direction since the last update, and updates 80 the position coordinates of the parcel unit to its current position.

[0019] After the location of the parcel unit has been updated, the multi-core classification processor 36 (FIG. 1) executes in parallel an in-target task 82 that determines when the parcel unit is in an object zone 84 within the image capture area 24 so that the camera 22 can be operated to capture an image of the parcel unit within the object zone. Before executing the in-target task 82, the classification processor 36 first limits execution of the in-target task 86 to parcel units that have not yet been imaged.

[0020] Each of the target targeting tasks 82 first determines whether a parcel unit is within a target zone 84 (FIG. 1) by comparing its coordinates with the coordinate range of the target zone 88. If the parcel unit is determined to be within the target zone, the task adds the parcel unit to a target list 90, otherwise it does nothing 92. The classification processor then checks the target list 94, as shown in FIG. 2. If one or more parcel units are newly added to the target list, the classification processor 36 signals the camera 22 (FIG. 1) to capture 96 an image of the parcel unit within the target zone 84. If there are no new parcel units in the target list, nothing is done 98 and the timer tick task 70 is completed.

[0021] With all the new images, the classification processor 36 executes in parallel the object processing tasks 100, which are shown in detail in FIG. 5. Each object processing task 100 includes an image cropping task 102 that first calculates 104 the relative pixel coordinates of the parcel units in the captured image. For example, the pixel coordinates of the corners of the parcel units can be used as the relative pixel coordinates. Based on those coordinates, the image cropping task 102 then performs a coordinate rotation 106 of the image containing the parcel units to orient the parcel units for effective cropping. The coordinate rotation results in a rotation of the coordinates of the parcel units in the captured image. The rotated captured image is then cropped 108 into smaller rectangular regions that encompass each of the rotated parcel unit images in the captured image. The pixels within the rectangular regions define cropped images, which are then stored 110.

[0022] For each cropped image, a classification task 112 is executed in parallel by the cores of the classification processor. Each cropped image contains a rectangular pixel array of one byte (0-255) RGB values. First, each classification task 112 pre-processes 114 the cropped image to bring it into a format for the particular classification program being used. For example, it may be necessary to resize the pixel array and pad it with zeros to fill the array. Examples of classifiers include: AlexNet; Inception Network v1, v2, v3, v3; MobileNet; PNASNet; SqueezeNet; ResNet, etc. The pixel array is fed as an input P to a neural network 116 (FIG. 6), which classifies the cropped image of the parcel unit 118 (FIG. 5). Once the parcel unit has been classified, the classification task stores 120 the classification in an infeed conveyor status table.

[0023] In this form, the classification processor uses artificial intelligence in the form of a neural network as a classifier. However, other artificial intelligence techniques, such as cascaded classifiers using Haar features, fully connected networks, convolutional neural networks, support vector machines, Bayesian neural networks, k-NN networks, Parzen neural networks, and fuzzy logic, may be used as classifiers to classify images of parcel units. Neural network 116 in FIG. 6 represents an exemplary neural network that illustrates how artificial intelligence is used to classify parcel units. The input P is the RGB value of a pixel in the pixel array of the preprocessed cropped image. The input is multiplied by a weighting factor w that is different for each line connecting the input P to a first layer neuron N1 in the hidden layer 122. Each first layer neuron N1 has an activation value that is the normalized sum of the products of the input values ​​P and the weights connected to it, plus a bias term. The normalization is achieved by a mathematical function that maps the sum to a limited range, for example from 0 to 1. In a similar manner, the activation values ​​of the first hidden layer N1 are applied to neurons N2 of the second hidden layer. The activation values ​​of neurons N2 of the second hidden layer are then used to calculate the outputs (A through E). An actual neural network used to classify parcel units could potentially have more than two hidden layers, with approximately six neurons shown for each hidden layer.

[0024] The outputs A-E represent a set of different classifications of parcel units. The classification task is to assign to each parcel unit the classification with the largest activation value among the five outputs. For example, A=plastic bags; B=recognized single parcel that is not a plastic bag; C=package stacks with less than a predefined percentage overlap (e.g., 25%); D=package stacks with more than a predefined percentage overlap; E=unrecognized parcels. Of course, other sets of parcel unit classifications can be used to identify other parcel types or parcel characteristics, such as surface textures, rips or tears, wet stains, particular colors, particular visible indicators, or even wrinkled plastic bags that have rendered machine-readable indicators such as bar codes illegible.

[0025] Neural network classifiers are trained by feeding the network a number of cropped images of parcel units that correspond to output classifications. The training adjusts the weights w and biases of the neurons in each layer to minimize the difference between a cost function, i.e., the desired output for an image (which is 0 for all outputs other than the one that corresponds to that image and 1 for the one that corresponds to that image), and the output computed by the neural network. The training process iteratively adjusts the weights and biases for each training image using conventional backpropagation techniques. Training is typically performed offline rather than in real time.

[0026] 5, the classification determined by the neural network for each parcel unit is sent to a control processor 124, and invalid parcel units and previously classified parcel units are removed from the table of items to be classified 126. Metadata such as the cropped image, timestamp, and pixel coordinates of the parcel unit are sent 128 to storage device 130 for offline analysis.

[0027] The control processor 34 in FIG. 1 checks the position of the parcel units 13 on the infeed conveyor section 12, updated by the sorting processor's parcel tracking task. Once the parcel units 13 reach the sorting conveyor section 25, the control processor 34 takes over calculating the position of each parcel unit and determining its destination and trajectory. The destination of each parcel unit depends on its classification. For example, unrecognized parcel units (classification E) and stacked parcels with an overlap of 25% or more (classification D) are sent to a destination 30 staffed by a human worker 32. It is also possible that all parcel units of any classification exceeding a predetermined maximum size can be sent to a human-staffed destination 30. In this format, their trajectory is straight in the conveying direction 14. Plastic bags (class A), recognized single parcels that are not plastic bags (class B), and stacked parcels with less than 25% overlap (class C) are diverted to one or the other of flanking conveyors 28, 29 which transport the parcels downstream for further processing. To control the trajectory 132 of each parcel unit 13, control processor 34 selectively activates roller belts or roller sections of a powered roller conveyor, or shoes of a shoe sorter, or pushers to divert the parcel unit to a designated destination.

[0028] The invention has been described in detail in an exemplary form using a computer processing system including two processors and a PLC. The system provides redundancy in the event of failure of the sorting processor. In the event of a failure, the control processor can sort the parcels based solely on the footprint of the parcel units without the added benefit of the parcel units being sorted. Also, in the event of a control processor failure, the PLC can simply sort the parcels until the control processor is restored, creating a balanced flow on two flanking conveyors staffed with temporary human operators. However, other forms are possible. For example, the control processor and sorting processor could be implemented by a single multi-core processor within the computer processing system. As another example, the sorting processor need not be a multi-core processor that executes task instructions in parallel on individual cores or threads. The sorting processor could be a single-core processor that executes one task instruction at a time.

[0029] In an alternative form, the camera's image capture area could be located along the infeed conveyor section upstream of the occupancy detector. The sorting processor could then be tasked with continuously capturing images and storing the captured images in an image table in the processor's volatile memory with a time stamp. The control processor could detect the occupancies with an occupancy detector downstream of the image capture area. Using information about the speed of the infeed conveyor and the distance from the image capture area to the occupancy detection location, the sorting processor could associate each occupancy with a parcel unit image contained in the captured image in the image table. The sorting processor could then rotate, crop, pre-process, and classify the parcel units as shown in FIG. 5 before they exit the infeed conveyor section and enter the sorting conveyor section.

[0030] It is also possible to use the imaging system as a footprint detector to detect the footprint or location of a parcel unit without using a separate dedicated footprint detector. In addition to performing the tasks necessary to classify parcel units from their images, the sorting processor may also perform the task of acting as a parcel unit detector by recognizing and tracking the location of individual parcel units as they advance along the infeed conveyor section so that they can be sorted to their assigned destinations.

Claims

1. an infeed conveyor section for conveying the separated parcel units downstream in a conveying direction; a parcel unit detector for detecting the parcel unit at a detection location along the infeed conveyor section; an imaging system for capturing images of the parcel units within a target zone of the infeed conveyor section; a computer processing system executing program instructions for tracking the position of each of said parcel units as they are conveyed on said infeed conveyor section, and for providing each of said parcel unit images as input to a classifier trained to recognize a set of parcel units, and for assigning to each of said parcel units a classification corresponding to one of the members of the set of parcel units that the classifier is trained to recognize; a downstream conveyor section that receives the parcel units from the infeed conveyor section; wherein said computer processing system executes program instructions for controlling said downstream conveyor sections to transport each parcel unit according to a classification of said parcel units.

2. The computer processing system: program instructions for calculating the position of each of said parcel units on said infeed conveyor section from their footprints detected as they pass said detection location as they advance in said conveying direction; program instructions for controlling the imaging system to capture images of one or more ones of the parcel units when the calculated positions of one or more of the parcel units are within the target zone on the infeed conveyor section to generate images of the one or more ones of the parcel units within the target zone; program instructions for cropping said image into one or more cropped images, each said cropped image corresponding to one of said one or more parcel units within said target zone; program instructions for classifying said cropped image into one of said classes; 2. The conveyor system of claim 1, further comprising:

3. 3. The conveyor system of claim 2, wherein the computer processing system calculates coordinates of a center of gravity of the detected occupancy area, the center of gravity defining a location of the parcel unit on the infeed conveyor section.

4. 3. The conveyor system of claim 2, wherein said computer processing system rotates said image to a non-oblique orientation relative to said conveying direction prior to cropping said image.

5. 3. The conveyor system of claim 2, wherein the computer processing system includes a multi-core processor having multiple cores that execute program instructions in parallel.

6. 2. The conveyor system of claim 1, wherein the computer processing system includes a first processor and a second processor that controls the downstream conveyor section to sort each of the parcel units to one of a plurality of destinations depending on the classification of the parcel unit.

7. 7. The conveyor system of claim 6, wherein the set of parcel units that the classifier is trained to recognize is: (a) plastic bags; (b) recognized single parcels that are not plastic bags; (c) parcel stacks having an overlap of less than a predetermined percentage; and (d) parcel stacks having an overlap of more than a predetermined percentage.

8. 8. The conveyor system of claim 7, wherein the second processor controls the sorter to sort parcel units classified as a stack of parcels having an overlap greater than the predetermined percentage or as an unrecognized parcel unit to a destination staffed by a human worker.

9. 8. The conveyor system of claim 7, wherein the second processor controls the sorter to sort parcel units classified as a recognized single parcel or as a stack of parcels with less than a predetermined percentage overlap to a downstream destination for further processing.

10. The conveyor system of claim 1 , wherein the parcel unit detector detects a footprint of the parcel unit.

11. 1. A processor-implemented method for sorting parcels conveyed on an infeed conveyor section, comprising: (a) detecting a multiple parcel unit advancing on said infeed conveyor section in a conveying direction at a conveying speed past a detection location; (b) calculating the position of each of the multiple parcel units detected as passing through the detection location on the infeed conveyor section as they advance along the conveyor; (c) imaging one or more of the multiple parcel units within an image capture area on the infeed conveyor section to generate an image of the one or more of the multiple parcel units within the capture area; (d) cropping the image into one or more cropped images, each cropped image corresponding to one of one or more multiple parcel units within the image capture area; (e) using artificial intelligence to classify the cropped images into a plurality of classifications by assigning one of the classifications to each of the one or more multiple parcel units; (f) sorting each of the one or more multiple parcel units to a destination according to the classification of the parcel units; The method includes:

12. 12. The method of claim 11, comprising performing steps (b)-(e) in parallel on a multi-core processor.

13. 13. The method of claim 12, comprising performing steps (a) and (f) in a separate control processor in communication with the multi-core processor.

14. 14. The method of claim 13, wherein the multi-core processor transmits a classification assigned to each of the one or more multiple parcel units to the control processor.

15. 12. The method of claim 11, wherein the parcel units are classified as: (a) plastic bags; (b) recognized single parcels that are not plastic bags; (c) stacks of parcels that overlap less than a predetermined percentage; (d) stacks of parcels that overlap more than a predetermined percentage; and (e) unrecognized parcel units.

16. 16. The method of claim 15, comprising sorting parcel units classified as parcel stacks having an overlap greater than a predetermined percentage or as unrecognized parcel units to destinations staffed by human workers.

17. 16. The method of claim 15, comprising sorting parcel units having a footprint greater than a predetermined footprint size to destinations staffed by human workers.

18. 16. The method of claim 15, comprising sorting parcel units classified as recognized single parcels or as stacks of parcels with less than a predetermined percentage overlap onto a conveyor that carries the parcel units for further processing.

19. The method of claim 11 , wherein the classification is performed by a neural network trained to recognize different types of parcel units.

20. The method of claim 11 , further comprising calculating coordinates of corners of the detected occupancy area.

21. 12. The method of claim 11, comprising calculating coordinates of a center of gravity of the detected occupancy area, the center of gravity defining a position of the parcel unit on the infeed conveyor section.

22. The method of claim 11 including rotating the image to a non-oblique orientation relative to the conveying direction prior to cropping the image.

23. The method of claim 11 , including pre-processing the cropped images in parallel to format the cropped images for classification prior to classifying the cropped images.

24. an infeed conveyor section for conveying parcel units downstream in a conveying direction at a conveying speed; an occupancy detector disposed at a detection location along the infeed conveyor section; a camera positioned along the infeed conveyor section downstream of the detection location to capture an image of a capture area on the infeed conveyor section; a sortation conveyor section that receives the parcel units from the infeed conveyor section and selectively sorts the parcel units to a plurality of destinations; A control processor comprising: program instructions for operating said footprint detector to detect the footprint of said parcel unit advancing in said conveyance direction past said detection location; program instructions for controlling the sorting conveyor section and the infeed conveyor section; a control processor executing a classification processor in communication with the control processor, comprising: program instructions for calculating the position of each of said parcel units on said infeed conveyor section from their footprints detected as they pass said detection location as they advance in said conveying direction; program instructions for controlling the camera to capture an image of one or more of the parcel units when a calculated position of the one or more of the parcel units is within a target zone of a capture area on the infeed conveyor section to generate an image of the one or more of the parcel units within the capture area; program instructions for cropping the image into one or more cropped images, each cropped image corresponding to one of the one or more parcel units within the capture area; program instructions for classifying, using artificial intelligence, the cropped images into a plurality of classifications by assigning one of the classifications to each of the one or more parcel units; a classification processor that executes wherein the control processor controls the sorting conveyor section to sort each of the one or more parcel units to a destination according to a classification of the parcel unit.

25. 25. The conveyor system of claim 24, wherein the classification processor classifies the cropped images in a neural network trained to recognize various parcel units and transmits the classification to the control processor.

26. 25. The conveyor system of claim 24, wherein the control processor includes a virtual or separate programmable logic controller that controls the sorter conveyor section and the infeed conveyor section.

27. 25. The conveyor system of claim 24, wherein the sorting processor is a multi-core processor having multiple cores that execute the program instructions in parallel.

Images